简体中文 | [English](./README_en.md)
# 基于Pytorch实现的声纹识别系统
本分支为1.1版本，如果要使用之前的1.0版本请在[1.0分支](https://github.com/yeyupiaoling/VoiceprintRecognition-Pytorch/tree/release/1.0.5)使用。本项目使用了EcapaTdnn、ResNetSE、ERes2Net、CAM++等多种先进的声纹识别模型，不排除以后会支持更多模型，同时本项目也支持了MelSpectrogram、Spectrogram、MFCC、Fbank等多种数据预处理方法，使用了ArcFace Loss，ArcFace loss：Additive Angular Margin Loss（加性角度间隔损失函数），对应项目中的AAMLoss，对特征向量和权重归一化，对θ加上角度间隔m，角度间隔比余弦间隔在对角度的影响更加直接，除此之外，还支持AMLoss、ARMLoss、CELoss等多种损失函数。
**本项目是如果对你有帮助，欢迎Star，避免之后需要找不到了。**
**欢迎大家扫码入知识星球或者QQ群讨论，知识星球里面提供项目的模型文件和博主其他相关项目的模型文件，也包括其他一些资源。**
<div align="center">
<img src="https://yeyupiaoling.cn/zsxq.png" alt="知识星球" width="400">
<img src="https://yeyupiaoling.cn/qq.png" alt="QQ群" width="400">
</div>
使用环境：
- Anaconda 3
- Python 3.11
- Pytorch 2.4.0
- Windows 11 or Ubuntu 22.04
# 目录
- [项目介绍](#基于Pytorch实现的声纹识别系统)
- [项目记录](#项目记录)
- [项目特性](#项目特性)
- [安装环境](#安装环境)
- [创建数据](#创建数据)
- [修改预处理方法（可选）](#修改预处理方法可选)
- [提取特征（可选）](#提取特征可选)
- [训练模型](#训练模型)
- [评估模型](#评估模型)
- [推理接口](#推理接口)
- [声纹对比](#声纹对比)
- [声纹识别](#声纹识别)
- [说话人日志（分离说话人）](#说话人日志分离说话人)
# 项目记录
1. 2024.10.12：发布1.1版本。
# 项目特性
1. 支持模型：EcapaTdnn、TDNN、Res2Net、ResNetSE、ERes2Net、CAM++
2. 支持池化层：AttentiveStatsPool(ASP)、SelfAttentivePooling(SAP)、TemporalStatisticsPooling(TSP)、TemporalAveragePooling(TAP)、TemporalStatsPool(TSTP)
3. 支持损失函数：AAMLoss、SphereFace2、AMLoss、ARMLoss、CELoss、SubCenterLoss、TripletAngularMarginLoss
4. 支持预处理方法：MelSpectrogram、Spectrogram、MFCC、Fbank、Wav2vec2.0、WavLM
5. 支持数据增强方法：语速增强、音量增强、噪声增强、混响增强、SpecAugment
**模型论文：**
- EcapaTdnn：[ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification](https://arxiv.org/abs/2005.07143v3)
- PANNS：[PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition](https://arxiv.org/abs/1912.10211v5)
- TDNN：[Prediction of speech intelligibility with DNN-based performance measures](https://arxiv.org/abs/2203.09148)
- Res2Net：[Res2Net: A New Multi-scale Backbone Architecture](https://arxiv.org/abs/1904.01169)
- ResNetSE：[Squeeze-and-Excitation Networks](https://arxiv.org/abs/1709.01507)
- CAMPPlus：[CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking](https://arxiv.org/abs/2303.00332v3)
- ERes2Net：[An Enhanced Res2Net with Local and Global Feature Fusion for Speaker Verification](https://arxiv.org/abs/2305.12838v1)
# 模型下载
### 训练CN-Celeb数据，共有2796个说话人。
| 模型 | Params(M) | 数据集 | train speakers | threshold | EER | MinDCF | 模型下载 |
|:----------:|:---------:|:----------------------------------:|:--------------:|:---------:|:-------:|:-------:|:--------:|
| ERes2NetV2 | 6.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.20089 | 0.08071 | 0.45705 | 加入知识星球获取 |
| ERes2Net | 6.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.20014 | 0.08132 | 0.45544 | 加入知识星球获取 |
| CAM++ | 6.8 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23323 | 0.08332 | 0.48536 | 加入知识星球获取 |
| ResNetSE | 7.8 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.19066 | 0.08544 | 0.49142 | 加入知识星球获取 |
| EcapaTdnn | 6.1 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23646 | 0.09259 | 0.51378 | 加入知识星球获取 |
| TDNN | 2.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23858 | 0.10825 | 0.59545 | 加入知识星球获取 |
| Res2Net | 5.0 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.19526 | 0.12436 | 0.65347 | 加入知识星球获取 |
| CAM++ | 6.8 | 更大数据集 | 2W+ | 0.33 | 0.07874 | 0.52524 | 加入知识星球获取 |
| ERes2Net | 55.1 | 其他数据集 | 20W+ | 0.36 | 0.02936 | 0.18355 | 加入知识星球获取 |
| ERes2NetV2 | 56.2 | 其他数据集 | 20W+ | 0.36 | 0.03847 | 0.24301 | 加入知识星球获取 |
| CAM++ | 6.8 | 其他数据集 | 20W+ | 0.29 | 0.04765 | 0.31436 | 加入知识星球获取 |
说明：
1. 评估的测试集为[CN-Celeb的测试集](https://aistudio.baidu.com/aistudio/datasetdetail/233361)，包含196个说话人。
2. 使用语速增强分类大小翻三倍`speed_perturb_3_class: True`。
3. 使用的预处理方法为`Fbank`，损失函数为`AAMLoss`。
4. 参数数量不包含了分类器的参数数量。
5. 使用了噪声增强和混响增强。
### 训练VoxCeleb1&2数据，共有7205个说话人。
| 模型 | Params(M) | 数据集 | train speakers | threshold | EER | MinDCF | 模型下载 |
|:----------:|:---------:|:-----------:|:--------------:|:---------:|:-------:|:-------:|:--------:|
| CAM++ | 6.8 | VoxCeleb1&2 | 7205 | 0.22504 | 0.02436 | 0.15543 | 加入知识星球获取 |
| EcapaTdnn | 6.1 | VoxCeleb1&2 | 7205 | 0.24877 | 0.02480 | 0.16188 | 加入知识星球获取 |
| ResNetSE | 7.8 | VoxCeleb1&2 | 7205 | 0.22567 | 0.03189 | 0.23040 | 加入知识星球获取 |
| TDNN | 2.6 | VoxCeleb1&2 | 7205 | 0.23834 | 0.03486 | 0.26792 | 加入知识星球获取 |
| Res2Net | 5.0 | VoxCeleb1&2 | 7205 | 0.19472 | 0.04370 | 0.40072 | 加入知识星球获取 |
| ERes2Net | 6.6 | VoxCeleb1&2 | 7205 | | | | 加入知识星球获取 |
| CAM++ | 6.8 | 更大数据集 | 2W+ | 0.28 | 0.03182 | 0.23731 | 加入知识星球获取 |
| ERes2Net | 55.1 | 其他数据集 | 20W+ | 0.53 | 0.08904 | 0.62130 | 加入知识星球获取 |
| ERes2NetV2 | 56.2 | 其他数据集 | 20W+ | 0.52 | 0.08649 | 0.64193 | 加入知识星球获取 |
| CAM++ | 6.8 | 其他数据集 | 20W+ | 0.49 | 0.10334 | 0.71200 | 加入知识星球获取 |
说明：
1. 评估的测试集为[VoxCeleb1&2的测试集](https://aistudio.baidu.com/aistudio/datasetdetail/255977)，包含158个说话人。
2. 使用语速增强分类大小翻三倍`speed_perturb_3_class: True`。
3. 使用的预处理方法为`Fbank`，损失函数为`AAMLoss`。
4. 参数数量不包含了分类器的参数数量。
### 预处理方法效果对比实验
| 预处理方法 | 数据集 | train speakers | threshold | EER | MinDCF | 模型下载 |
|:--------------------------------------------------------------------------------:|:--------:|:--------------:|:---------:|:-------:|:-------:|:--------:|
| Fbank | CN-Celeb | 2796 | 0.14574 | 0.10988 | 0.58955 | 加入知识星球获取 |
| MFCC | CN-Celeb | 2796 | 0.14868 | 0.11483 | 0.61275 | 加入知识星球获取 |
| Spectrogram | CN-Celeb | 2796 | 0.14962 | 0.11613 | 0.60057 | 加入知识星球获取 |
| MelSpectrogram | CN-Celeb | 2796 | 0.13458 | 0.12498 | 0.60741 | 加入知识星球获取 |
| [wavlm-base-plus](https://huggingface.co/microsoft/wavlm-base-plus) | CN-Celeb | 2796 | 0.14166 | 0.13247 | 0.62451 | 加入知识星球获取 |
| [w2v-bert-2.0](https://huggingface.co/facebook/w2v-bert-2.0) | CN-Celeb | 2796 | | | | 加入知识星球获取 |
| [wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) | CN-Celeb | 2796 | | | | 加入知识星球获取 |
| [wavlm-large](https://huggingface.co/microsoft/wavlm-large) | CN-Celeb | 2796 | | | | 加入知识星球获取 |
说明：
1. 评估的测试集为[CN-Celeb的测试集](https://aistudio.baidu.com/aistudio/datasetdetail/233361)，包含196个说话人。
2. 实验数据为[CN-Celeb](http://openslr.org/82/)，实验模型为`CAM++`，损失函数为`AAMLoss`。
3. 数据使用`extract_features.py`提前提取特征，也就是说训练中没有使用对音频的数据增强。
4. `w2v-bert-2.0`、`wav2vec2-large-xlsr-53`是多语言数据预训练得到的，`wavlm-base-plus`、`wavlm-large`的预训练数据仅用英文。
### 损失函数效果对比实验
| 损失函数 | 数据集 | train speakers | threshold | EER | MinDCF | 模型下载 |
|:------------------------:|:--------:|:--------------:|:---------:|:-------:|:-------:|:--------:|
| AAMLoss | CN-Celeb | 2796 | 0.14574 | 0.10988 | 0.58955 | 加入知识星球获取 |
| SphereFace2 | CN-Celeb | 2796 | 0.20377 | 0.11309 | 0.61536 | 加入知识星球获取 |
| TripletAngularMarginLoss | CN-Celeb | 2796 | 0.28940 | 0.11749 | 0.63735 | 加入知识星球获取 |
| SubCenterLoss | CN-Celeb | 2796 | 0.13126 | 0.11775 | 0.56995 | 加入知识星球获取 |
| ARMLoss | CN-Celeb | 2796 | 0.14563 | 0.11805 | 0.57171 | 加入知识星球获取 |
| AMLoss | CN-Celeb | 2796 | 0.12870 | 0.12301 | 0.63263 | 加入知识星球获取 |
| CELoss | CN-Celeb | 2796 | 0.13607 | 0.12684 | 0.65176 | 加入知识星球获取 |
说明：
1. 评估的测试集为[CN-Celeb的测试集](https://aistudio.baidu.com/aistudio/datasetdetail/233361)，包含196个说话人。
2. 实验数据为[CN-Celeb](http://openslr.org/82/)，实验模型为`CAM++`，预处理方法为`Fbank`。
3. 数据使用`extract_features.py`提前提取特征，也就是说训练中没有使用对音频的数据增强。
## 安装环境
- 首先安装的是Pytorch的GPU版本，如果已经安装过了，请跳过。
```shell
conda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=11.8 -c pytorch -c nvidia
```
- 安装ppvector库。
使用pip安装，命令如下：
```shell
python -m pip install mvector -U -i https://pypi.tuna.tsinghua.edu.cn/simple
```
**建议源码安装**，源码安装能保证使用最新代码。
```shell
git clone https://github.com/yeyupiaoling/VoiceprintRecognition-Pytorch.git
cd VoiceprintRecognition-Pytorch/
pip install .
```
# 创建数据
本教程笔者使用的是[CN-Celeb](https://openslr.elda.org/resources/82)，这个数据集一共有约3000个人的语音数据，有65W+条语音数据，下载之后要解压数据集到`dataset`目录，另外如果要评估，还需要下载[CN-Celeb的测试集](https://aistudio.baidu.com/aistudio/datasetdetail/233361)。如果读者有其他更好的数据集，可以混合在一起使用，但最好是要用python的工具模块aukit处理音频，降噪和去除静音。
首先是创建一个数据列表，数据列表的格式为`<语音文件路径\t语音分类标签>`，创建这个列表主要是方便之后的读取，也是方便读取使用其他的语音数据集，语音分类标签是指说话人的唯一ID，不同的语音数据集，可以通过编写对应的生成数据列表的函数，把这些数据集都写在同一个数据列表中。
执行`create_data.py`程序完成数据准备。
```shell
python create_data.py
```
执行上面的程序之后，会生成以下的数据格式，如果要自定义数据，参考如下数据列表，前面是音频的相对路径，后面的是该音频对应的说话人的标签，就跟分类一样。**自定义数据集的注意**，测试数据列表的ID可以不用跟训练的ID一样，也就是说测试的数据的说话人可以不用出现在训练集，只要保证测试数据列表中同一个人相同的ID即可。
```
dataset/CN-Celeb2_flac/data/id11999/recitation-03-019.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-10-023.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-06-025.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-04-014.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-06-030.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-10-032.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-06-028.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-10-031.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-05-003.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-04-017.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-10-016.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-09-001.flac 2795
dataset/CN-Celeb2_flac/data/id11999/recitation-05-010.flac 2795
```
# 修改预处理方法（可选）
配置文件中默认使用的是Fbank预处理方法，如果要使用其他预处理方法，可以修改配置文件中的安装下面方式修改，具体的值可以根据自己情况修改。如果不清楚如何设置参数，可以直接删除该部分，直接使用默认值。
```yaml
# 数据预处理参数
preprocess_conf:
# 是否使用HF上的Wav2Vec2类似模型提取音频特征
use_hf_model: False
# 音频预处理方法，也可以叫特征提取方法
# 当use_hf_model为False时，支持：MelSpectrogram、Spectrogram、MFCC、Fbank
# 当use_hf_model为True时，指定的是HuggingFace的模型或者本地路径，比如facebook/w2v-bert-2.0或者./feature_models/w2v-bert-2.0
feature_method: 'Fbank'
# 当use_hf_model为False时，设置API参数，更参数查看对应API，不清楚的可以直接删除该部分，直接使用默认值。
# 当use_hf_model为True时，可以设置参数use_gpu，指定是否使用GPU提取特征
method_args:
sample_frequency: 16000
num_mel_bins: 80
```
# 提取特征（可选）
在训练过程中，首先是要读取音频数据，然后提取特征，最后再进行训练。其中读取音频数据、提取特征也是比较消耗时间的，所以我们可以选择提前提取好取特征，训练模型的是就可以直接加载提取好的特征，这样训练速度会更快。这个提取特征是可选择，如果没有提取好的特征，训练模型的时候就会从读取音频数据，然后提取特征开始。提取特征步骤如下：
1. 执行`extract_features.py`，提取特征，特征会保存在`dataset/features`目录下，并生成新的数据列表`train_list_features.txt`、`enroll_list_features.txt`和`trials_list_features.txt`。
```shell
python extract_features.py --configs=configs/cam++.yml --save_dir=dataset/features
```
2. 修改配置文件，将`dataset_conf.train_list`、`dataset_conf.enroll_list`和`dataset_conf.trials_list`修改为`train_list_features.txt`、`enroll_list_features.txt`和`trials_list_features.txt`。
# 训练模型
使用`train.py`训练模型，本项目支持多个音频预处理方式，通过`configs/ecapa_tdnn.yml`配置文件的参数`preprocess_conf.feature_method`可以指定，`MelSpectrogram`为梅尔频谱，`Spectrogram`为语谱图，`MFCC`梅尔频谱倒谱系数等等。通过参数`augment_conf_path`可以指定数据增强方式。训练过程中，会使用VisualDL保存训练日志，通过启动VisualDL可以随时查看训练结果，启动命令`visualdl --logdir=log --host 0.0.0.0`
```shell
# 单卡训练
CUDA_VISIBLE_DEVICES=0 python train.py
# 多卡训练
CUDA_VISIBLE_DEVICES=0,1 torchrun --standalone --nnodes=1 --nproc_per_node=2 train.py
```
训练输出日志：
```
[2023-08-05 09:52:06.497988 INFO ] utils:print_arguments:13 - ----------- 额外配置参数 -----------
[2023-08-05 09:52:06.498094 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml
[2023-08-05 09:52:06.498149 INFO ] utils:print_arguments:15 - do_eval: True
[2023-08-05 09:52:06.498191 INFO ] utils:print_arguments:15 - local_rank: 0
[2023-08-05 09:52:06.498230 INFO ] utils:print_arguments:15 - pretrained_model: None
[2023-08-05 09:52:06.498269 INFO ] utils:print_arguments:15 - resume_model: None
[2023-08-05 09:52:06.498306 INFO ] utils:print_arguments:15 - save_model_path: models/
[2023-08-05 09:52:06.498342 INFO ] utils:print_arguments:15 - use_gpu: True
[2023-08-05 09:52:06.498378 INFO ] utils:print_arguments:16 - ------------------------------------------------
[2023-08-05 09:52:06.513761 INFO ] utils:print_arguments:18 - ----------- 配置文件参数 -----------
[2023-08-05 09:52:06.513906 INFO ] utils:print_arguments:21 - dataset_conf:
[2023-08-05 09:52:06.513957 INFO ] utils:print_arguments:24 - dataLoader:
[2023-08-05 09:52:06.513995 INFO ] utils:print_arguments:26 - batch_size: 64
[2023-08-05 09:52:06.514031 INFO ] utils:print_arguments:26 - num_workers: 4
[2023-08-05 09:52:06.514066 INFO ] utils:print_arguments:28 - do_vad: False
[2023-08-05 09:52:06.514101 INFO ] utils:print_arguments:28 - enroll_list: dataset/enroll_list.txt
[2023-08-05 09:52:06.514135 INFO ] utils:print_arguments:24 - eval_conf:
[2023-08-05 09:52:06.514169 INFO ] utils:print_arguments:26 - batch_size: 1
[2023-08-05 09:52:06.514203 INFO ] utils:print_arguments:26 - max_duration: 20
[2023-08-05 09:52:06.514237 INFO ] utils:print_arguments:28 - max_duration: 3
[2023-08-05 09:52:06.514274 INFO ] utils:print_arguments:28 - min_duration: 0.5
[2023-08-05 09:52:06.514308 INFO ] utils:print_arguments:28 - noise_aug_prob: 0.2
[2023-08-05 09:52:06.514342 INFO ] utils:print_arguments:28 - noise_dir: dataset/noise
[2023-08-05 09:52:06.514374 INFO ] utils:print_arguments:28 - num_speakers: 3242
[2023-08-05 09:52:06.514408 INFO ] utils:print_arguments:28 - sample_rate: 16000
[2023-08-05 09:52:06.514441 INFO ] utils:print_arguments:28 - speed_perturb: True
[2023-08-05 09:52:06.514475 INFO ] utils:print_arguments:28 - target_dB: -20
[2023-08-05 09:52:06.514508 INFO ] utils:print_arguments:28 - train_list: dataset/train_list.txt
[2023-08-05 09:52:06.514542 INFO ] utils:print_arguments:28 - trials_list: dataset/trials_list.txt
[2023-08-05 09:52:06.514575 INFO ] utils:print_arguments:28 - use_dB_normalization: True
[2023-08-05 09:52:06.514609 INFO ] utils:print_arguments:21 - loss_conf:
[2023-08-05 09:52:06.514643 INFO ] utils:print_arguments:24 - args:
[2023-08-05 09:52:06.514678 INFO ] utils:print_arguments:26 - easy_margin: False
[2023-08-05 09:52:06.514713 INFO ] utils:print_arguments:26 - margin: 0.2
[2023-08-05 09:52:06.514746 INFO ] utils:print_arguments:26 - scale: 32
[2023-08-05 09:52:06.514779 INFO ] utils:print_arguments:24 - margin_scheduler_args:
[2023-08-05 09:52:06.514814 INFO ] utils:print_arguments:26 - final_margin: 0.3
[2023-08-05 09:52:06.514848 INFO ] utils:print_arguments:28 - use_loss: AAMLoss
[2023-08-05 09:52:06.514882 INFO ] utils:print_arguments:28 - use_margin_scheduler: True
[2023-08-05 09:52:06.514915 INFO ] utils:print_arguments:21 - model_conf:
[2023-08-05 09:52:06.514950 INFO ] utils:print_arguments:24 - backbone:
[2023-08-05 09:52:06.514984 INFO ] utils:print_arguments:26 - embd_dim: 192
[2023-08-05 09:52:06.515017 INFO ] utils:print_arguments:26 - pooling_type: ASP
[2023-08-05 09:52:06.515050 INFO ] utils:print_arguments:24 - classifier:
[2023-08-05 09:52:06.515084 INFO ] utils:print_arguments:26 - num_blocks: 0
[2023-08-05 09:52:06.515118 INFO ] utils:print_arguments:21 - optimizer_conf:
[2023-08-05 09:52:06.515154 INFO ] utils:print_arguments:28 - learning_rate: 0.001
[2023-08-05 09:52:06.515188 INFO ] utils:print_arguments:28 - optimizer: Adam
[2023-08-05 09:52:06.515221 INFO ] utils:print_arguments:28 - scheduler: CosineAnnealingLR
[2023-08-05 09:52:06.515254 INFO ] utils:print_arguments:28 - scheduler_args: None
[2023-08-05 09:52:06.515289 INFO ] utils:print_arguments:28 - weight_decay: 1e-06
[2023-08-05 09:52:06.515323 INFO ] utils:print_arguments:21 - preprocess_conf:
[2023-08-05 09:52:06.515357 INFO ] utils:print_arguments:28 - feature_method: MelSpectrogram
[2023-08-05 09:52:06.515390 INFO ] utils:print_arguments:24 - method_args:
[2023-08-05 09:52:06.515426 INFO ] utils:print_arguments:26 - f_max: 14000.0
[2023-08-05 09:52:06.515460 INFO ] utils:print_arguments:26 - f_min: 50.0
[2023-08-05 09:52:06.515493 INFO ] utils:print_arguments:26 - hop_length: 320
[2023-08-05 09:52:06.515527 INFO ] utils:print_arguments:26 - n_fft: 1024
[2023-08-05 09:52:06.515560 INFO ] utils:print_arguments:26 - n_mels: 64
[2023-08-05 09:52:06.515593 INFO ] utils:print_arguments:26 - sample_rate: 16000
[2023-08-05 09:52:06.515626 INFO ] utils:print_arguments:26 - win_length: 1024
[2023-08-05 09:52:06.515660 INFO ] utils:print_arguments:21 - train_conf:
[2023-08-05 09:52:06.515694 INFO ] utils:print_arguments:28 - log_interval: 100
[2023-08-05 09:52:06.515728 INFO ] utils:print_arguments:28 - max_epoch: 30
[2023-08-05 09:52:06.515761 INFO ] utils:print_arguments:30 - use_model: EcapaTdnn
[2023-08-05 09:52:06.515794 INFO ] utils:print_arguments:31 - ------------------------------------------------
······
===============================================================================================
Layer (type:depth-idx) Output Shape Param #
===============================================================================================
Sequential [1, 9726] --
├─EcapaTdnn: 1-1 [1, 192] --
│ └─Conv1dReluBn: 2-1 [1, 512, 98] --
│ │ └─Conv1d: 3-1 [1, 512, 98] 163,840
│ │ └─BatchNorm1d: 3-2 [1, 512, 98] 1,024
│ └─Sequential: 2-2 [1, 512, 98] --
│ │ └─Conv1dReluBn: 3-3 [1, 512, 98] 263,168
│ │ └─Res2Conv1dReluBn: 3-4 [1, 512, 98] 86,912
│ │ └─Conv1dReluBn: 3-5 [1, 512, 98] 263,168
│ │ └─SE_Connect: 3-6 [1, 512, 98] 262,912
│ └─Sequential: 2-3 [1, 512, 98] --
│ │ └─Conv1dReluBn: 3-7 [1, 512, 98] 263,168
│ │ └─Res2Conv1dReluBn: 3-8 [1, 512, 98] 86,912
│ │ └─Conv1dReluBn: 3-9 [1, 512, 98] 263,168
│ │ └─SE_Connect: 3-10 [1, 512, 98] 262,912
│ └─Sequential: 2-4 [1, 512, 98] --
│ │ └─Conv1dReluBn: 3-11 [1, 512, 98] 263,168
│ │ └─Res2Conv1dReluBn: 3-12 [1, 512, 98] 86,912
│ │ └─Conv1dReluBn: 3-13 [1, 512, 98] 263,168
│ │ └─SE_Connect: 3-14 [1, 512, 98] 262,912
│ └─Conv1d: 2-5 [1, 1536, 98] 2,360,832
│ └─AttentiveStatsPool: 2-6 [1, 3072] --
│ │ └─Conv1d: 3-15 [1, 128, 98] 196,736
│ │ └─Conv1d: 3-16 [1, 1536, 98] 198,144
│ └─BatchNorm1d: 2-7 [1, 3072] 6,144
│ └─Linear: 2-8 [1, 192] 590,016
│ └─BatchNorm1d: 2-9 [1, 192] 384
├─SpeakerIdentification: 1-2 [1, 9726] 1,867,392
===============================================================================================
Total params: 8,012,992
Trainable params: 8,012,992
Non-trainable params: 0
Total mult-adds (M): 468.81
===============================================================================================
Input size (MB): 0.03
Forward/backward pass size (MB): 10.36
Params size (MB): 32.05
Estimated Total Size (MB): 42.44
===============================================================================================
[2023-08-05 09:52:08.084231 INFO ] trainer:train:388 - 训练数据：874175
[2023-08-05 09:52:09.186542 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [0/13659], loss: 11.95824, accuracy: 0.00000, learning rate: 0.00100000, speed: 58.09 data/sec, eta: 5 days, 5:24:08
[2023-08-05 09:52:22.477905 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [100/13659], loss: 10.35675, accuracy: 0.00278, learning rate: 0.00100000, speed: 481.65 data/sec, eta: 15:07:15
[2023-08-05 09:52:35.948581 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [200/13659], loss: 10.22089, accuracy: 0.00505, learning rate: 0.00100000, speed: 475.27 data/sec, eta: 15:19:12
[2023-08-05 09:52:49.249098 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [300/13659], loss: 10.00268, accuracy: 0.00706, learning rate: 0.00100000, speed: 481.45 data/sec, eta: 15:07:11
[2023-08-05 09:53:03.716015 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [400/13659], loss: 9.76052, accuracy: 0.00830, learning rate: 0.00100000, speed: 442.74 data/sec, eta: 16:26:16
[2023-08-05 09:53:18.258807 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [500/13659], loss: 9.50189, accuracy: 0.01060, learning rate: 0.00100000, speed: 440.46 data/sec, eta: 16:31:08
[2023-08-05 09:53:31.618354 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [600/13659], loss: 9.26083, accuracy: 0.01256, learning rate: 0.00100000, speed: 479.50 data/sec, eta: 15:10:12
[2023-08-05 09:53:45.439642 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [700/13659], loss: 9.03548, accuracy: 0.01449, learning rate: 0.00099999, speed: 463.63 data/sec, eta: 15:41:08
```
启动VisualDL：`visualdl --logdir=log --host 0.0.0.0`，VisualDL页面如下：
<div align="center">
<img src="./docs/images/log.jpg" alt="VisualDL页面" width="600">
</div>
# 评估模型
训练结束之后会保存预测模型，我们用预测模型来预测测试集中的音频特征，然后使用音频特征进行两两对比，计算EER和MinDCF。
```shell
python eval.py
```
输出类似如下：
```
······
------------------------------------------------
W0425 08:27:32.057426 17654 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2
W0425 08:27:32.065165 17654 device_context.cc:465] device: 0, cuDNN Version: 7.6.
[2023-03-16 20:20:47.195908 INFO ] trainer:evaluate:341 - 成功加载模型：models/EcapaTdnn_Fbank/best_model/model.pth
100%|███████████████████████████| 84/84 [00:28<00:00, 2.95it/s]
开始两两对比音频特征...
100%|███████████████████████████| 5332/5332 [00:05<00:00, 1027.83it/s]
评估消耗时间：65s，threshold：0.26，EER: 0.14739, MinDCF: 0.41999
```
# 推理接口
下面给出了几个常用的接口，更多接口请参考`mvector/predict.py`，也可以往下看`声纹对比`和`声纹识别`的例子。
```python
from mvector.predict import MVectorPredictor
predictor = MVectorPredictor(configs='configs/cam++.yml',
model_path='models/CAMPPlus_Fbank/best_model/')
# 获取音频特征
embedding = predictor.predict(audio_data='dataset/a_1.wav')
# 获取两个音频的相似度
similarity = predictor.contrast(audio_data1='dataset/a_1.wav', audio_data2='dataset/a_2.wav')
# 注册用户音频
predictor.register(user_name='夜雨飘零', audio_data='dataset/test.wav')
# 识别用户音频
name, score = predictor.recognition(audio_data='dataset/test1.wav')
# 获取所有用户
users_name = predictor.get_users()
# 删除用户音频
predictor.remove_user(user_name='夜雨飘零')
```
# 声纹对比
下面开始实现声纹对比，创建`infer_contrast.py`程序，首先介绍几个重要的函数，`predict()`函数是可以获取声纹特征，`predict_batch()`函数是可以获取一批的声纹特征，`contrast()`函数可以对比两条音频的相似度，`register()`函数注册一条音频到声纹库里面，`recognition()`函输入一条音频并且从声纹库里面对比识别，`remove_user()`函数移除你好。声纹库里面的注册人。我们输入两个语音，通过预测函数获取他们的特征数据，使用这个特征数据可以求他们的对角余弦值，得到的结果可以作为他们相识度。对于这个相识度的阈值`threshold`，读者可以根据自己项目的准确度要求进行修改。
```shell
python infer_contrast.py --audio_path1=audio/a_1.wav --audio_path2=audio/b_2.wav
```
输出类似如下：
```
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:13 - ----------- 额外配置参数 -----------
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - audio_path1: dataset/a_1.wav
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - audio_path2: dataset/b_2.wav
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - model_path: models/EcapaTdnn_Fbank/best_model/
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - threshold: 0.6
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - use_gpu: True
[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:16 - ------------------------------------------------
······································································
W0425 08:29:10.006249 21121 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2
W0425 08:29:10.008555 21121 device_context.cc:465] device: 0, cuDNN Version: 7.6.
成功加载模型参数和优化方法参数：models/EcapaTdnn_Fbank/best_model/model.pth
audio/a_1.wav 和 audio/b_2.wav 不是同一个人，相似度为：-0.09565544128417969
```
同时还提供了有GUI界面的声纹对比程序，执行`infer_contrast_gui.py`启动程序，界面如下，分别选择两个音频，点击开始判断，就可以判断它们是否是同一个人。
<div align="center">
<img src="./docs/images/contrast.jpg" alt="声纹对比界面">
</div>
# 声纹识别
在新闻识别里面主要使用到`register()`函数和`recognition()`函数，首先使用`register()`函数函数来注册音频到声纹库里面，也可以直接把文件添加到`audio_db`文件夹里面，使用的时候通过`recognition()`函数来发起识别，输入一条音频，就可以从声纹库里面识别到所需要的说话人。
有了上面的声纹识别的函数，读者可以根据自己项目的需求完成声纹识别的方式，例如笔者下面提供的是通过录音来完成声纹识别。首先必须要加载语音库中的语音，语音库文件夹为`audio_db`，然后用户回车后录音3秒钟，然后程序会自动录音，并使用录音到的音频进行声纹识别，去匹配语音库中的语音，获取用户的信息。通过这样方式，读者也可以修改成通过服务请求的方式完成声纹识别，例如提供一个API供APP调用，用户在APP上通过声纹登录时，把录音到的语音发送到后端完成声纹识别，再把结果返回给APP，前提是用户已经使用语音注册，并成功把语音数据存放在`audio_db`文件夹中。
```shell
python infer_recognition.py
```
输出类似如下：
```
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:13 - ----------- 额外配置参数 -----------
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - audio_db_path: audio_db/
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - model_path: models/EcapaTdnn_Fbank/best_model/
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - record_seconds: 3
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - threshold: 0.6
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - use_gpu: True
[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:16 - ------------------------------------------------
······································································
W0425 08:30:13.257884 23889 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2
W0425 08:30:13.260191 23889 device_context.cc:465] device: 0, cuDNN Version: 7.6.
成功加载模型参数和优化方法参数：models/ecapa_tdnn/model.pth
Loaded 沙瑞金 audio.
Loaded 李达康 audio.
请选择功能，0为注册音频到声纹库，1为执行声纹识别：0
按下回车键开机录音，录音3秒中：
开始录音......
录音已结束!
请输入该音频用户的名称：夜雨飘零
请选择功能，0为注册音频到声纹库，1为执行声纹识别：1
按下回车键开机录音，录音3秒中：
开始录音......
录音已结束!
识别说话的为：夜雨飘零，相似度为：0.920434
```
同时还提供了有GUI界面的声纹识别程序，执行`infer_recognition_gui.py`启动，点击`注册音频到声纹库`按钮，理解开始说话，录制3秒钟，然后输入注册人的名称，之后可以`执行声纹识别`按钮，然后立即说话，录制3秒钟后，等待识别结果。`删除用户`按钮可以删除用户。`实时识别`按钮可以实时识别，可以一直录音，一直识别。
<div align="center">
<img src="./docs/images/recognition.jpg" alt="声纹识别界面">
</div>
# 说话人日志（分离说话人）
执行`infer_speaker_diarization.py`程序，输入音频路径，就可以分离出说话人，并显示结果，建议音频长度不要低于10秒。更多功能可以查看该程序参数。
```shell
python infer_speaker_diarization.py --audio_path=dataset/test_long.wav
```
输出类似如下：
```
2024-10-10 19:30:40.768 | INFO | mvector.predict:__init__:61 - 成功加载模型参数：models/CAMPPlus_Fbank/best_model/model.pth
2024-10-10 19:30:40.795 | INFO | mvector.predict:__create_index:127 - 声纹特征索引创建完成，一共有3个用户，分别是：['沙瑞金', '夜雨飘零', '李达康']
2024-10-10 19:30:40.796 | INFO | mvector.predict:__load_audio_db:142 - 正在加载声纹库数据...
100%|██████████| 3/3 [00:00<?, ?it/s]
2024-10-10 19:30:40.798 | INFO | mvector.predict:__create_index:127 - 声纹特征索引创建完成，一共有3个用户，分别是：['沙瑞金', '夜雨飘零', '李达康']
2024-10-10 19:30:40.798 | INFO | mvector.predict:__load_audio_db:172 - 声纹库数据加载完成！
识别结果：
{'speaker': '沙瑞金', 'start': 0.0, 'end': 2.0}
{'speaker': '陌生人1', 'start': 4.0, 'end': 7.0}
{'speaker': '李达康', 'start': 7.0, 'end': 8.0}
{'speaker': '沙瑞金', 'start': 9.0, 'end': 12.0}
{'speaker': '沙瑞金', 'start': 13.0, 'end': 14.0}
{'speaker': '陌生人1', 'start': 15.0, 'end': 19.0}
```
显示结果图像如下，可以通过`空格`键控制播放音频，点击位置可以跳转音频到指定位置：
<div align="center">
<img src="./docs/images/speaker_diarization.jpg" alt="说话人日志" width="800">
</div>
项目同样提供了GUI界面的程序，执行`infer_speaker_diarization_gui.py`程序。更多功能可以查看该程序参数。
```shell
python infer_speaker_diarization_gui.py
```
可以打开这样一个页面，进行说话人识别：
<div align="center">
<img src="./docs/images/speaker_diarization_gui.png" alt="说话人日志" width="800">
</div>
注意：如果说话人名字是中文的，需要设置安装字体才能正常显示，一般情况下Windows无需安装，Ubuntu需要安装。如果Windows确实是缺少字体，只需要[字体文件](https://github.com/tracyone/program_font)这里下载`.ttf`格式的文件，复制到`C:\Windows\Fonts`即可。Ubuntu系统操作如下。
1. 安装字体
```shell
git clone https://github.com/tracyone/program_font && cd program_font && ./install.sh
```
2. 执行下面Python代码
```python
import matplotlib
import shutil
import os
path = matplotlib.matplotlib_fname()
path = path.replace('matplotlibrc', 'fonts/ttf/')
print(path)
shutil.copy('/usr/share/fonts/MyFonts/simhei.ttf', path)
user_dir = os.path.expanduser('~')
shutil.rmtree(f'{user_dir}/.cache/matplotlib', ignore_errors=True)
```
# 其他版本
- Tensorflow：[VoiceprintRecognition-Tensorflow](https://github.com/yeyupiaoling/VoiceprintRecognition-Tensorflow)
- PaddlePaddle：[VoiceprintRecognition-PaddlePaddle](https://github.com/yeyupiaoling/VoiceprintRecognition-PaddlePaddle)
- Keras：[VoiceprintRecognition-Keras](https://github.com/yeyupiaoling/VoiceprintRecognition-Keras)
## 打赏作者
<br/>
<div align="center">
<p>打赏一块钱支持一下作者</p>
<img src="https://yeyupiaoling.cn/reward.png" alt="打赏作者" width="400">
</div>
# 参考资料
1. https://github.com/PaddlePaddle/PaddleSpeech
2. https://github.com/yeyupiaoling/PaddlePaddle-MobileFaceNets
3. https://github.com/yeyupiaoling/PPASR
4. https://github.com/alibaba-damo-academy/3D-Speaker
5. https://github.com/wenet-e2e/wespeaker
Raw data
{
"_id": null,
"home_page": "https://github.com/yeyupiaoling/VoiceprintRecognition_Pytorch",
"name": "mvector",
"maintainer": null,
"docs_url": null,
"requires_python": null,
"maintainer_email": null,
"keywords": "Voice, Pytorch",
"author": "yeyupiaoling",
"author_email": null,
"download_url": "https://github.com/yeyupiaoling/VoiceprintRecognition_Pytorch.git",
"platform": null,
"description": "\u7b80\u4f53\u4e2d\u6587 | [English](./README_en.md)\r\n\r\n# \u57fa\u4e8ePytorch\u5b9e\u73b0\u7684\u58f0\u7eb9\u8bc6\u522b\u7cfb\u7edf\r\n\r\n\r\n\r\n\r\n\r\n\r\n\r\n\u672c\u5206\u652f\u4e3a1.1\u7248\u672c\uff0c\u5982\u679c\u8981\u4f7f\u7528\u4e4b\u524d\u76841.0\u7248\u672c\u8bf7\u5728[1.0\u5206\u652f](https://github.com/yeyupiaoling/VoiceprintRecognition-Pytorch/tree/release/1.0.5)\u4f7f\u7528\u3002\u672c\u9879\u76ee\u4f7f\u7528\u4e86EcapaTdnn\u3001ResNetSE\u3001ERes2Net\u3001CAM++\u7b49\u591a\u79cd\u5148\u8fdb\u7684\u58f0\u7eb9\u8bc6\u522b\u6a21\u578b\uff0c\u4e0d\u6392\u9664\u4ee5\u540e\u4f1a\u652f\u6301\u66f4\u591a\u6a21\u578b\uff0c\u540c\u65f6\u672c\u9879\u76ee\u4e5f\u652f\u6301\u4e86MelSpectrogram\u3001Spectrogram\u3001MFCC\u3001Fbank\u7b49\u591a\u79cd\u6570\u636e\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u4f7f\u7528\u4e86ArcFace Loss\uff0cArcFace loss\uff1aAdditive Angular Margin Loss\uff08\u52a0\u6027\u89d2\u5ea6\u95f4\u9694\u635f\u5931\u51fd\u6570\uff09\uff0c\u5bf9\u5e94\u9879\u76ee\u4e2d\u7684AAMLoss\uff0c\u5bf9\u7279\u5f81\u5411\u91cf\u548c\u6743\u91cd\u5f52\u4e00\u5316\uff0c\u5bf9\u03b8\u52a0\u4e0a\u89d2\u5ea6\u95f4\u9694m\uff0c\u89d2\u5ea6\u95f4\u9694\u6bd4\u4f59\u5f26\u95f4\u9694\u5728\u5bf9\u89d2\u5ea6\u7684\u5f71\u54cd\u66f4\u52a0\u76f4\u63a5\uff0c\u9664\u6b64\u4e4b\u5916\uff0c\u8fd8\u652f\u6301AMLoss\u3001ARMLoss\u3001CELoss\u7b49\u591a\u79cd\u635f\u5931\u51fd\u6570\u3002\r\n\r\n**\u672c\u9879\u76ee\u662f\u5982\u679c\u5bf9\u4f60\u6709\u5e2e\u52a9\uff0c\u6b22\u8fceStar\uff0c\u907f\u514d\u4e4b\u540e\u9700\u8981\u627e\u4e0d\u5230\u4e86\u3002**\r\n\r\n**\u6b22\u8fce\u5927\u5bb6\u626b\u7801\u5165\u77e5\u8bc6\u661f\u7403\u6216\u8005QQ\u7fa4\u8ba8\u8bba\uff0c\u77e5\u8bc6\u661f\u7403\u91cc\u9762\u63d0\u4f9b\u9879\u76ee\u7684\u6a21\u578b\u6587\u4ef6\u548c\u535a\u4e3b\u5176\u4ed6\u76f8\u5173\u9879\u76ee\u7684\u6a21\u578b\u6587\u4ef6\uff0c\u4e5f\u5305\u62ec\u5176\u4ed6\u4e00\u4e9b\u8d44\u6e90\u3002**\r\n\r\n<div align=\"center\">\r\n <img src=\"https://yeyupiaoling.cn/zsxq.png\" alt=\"\u77e5\u8bc6\u661f\u7403\" width=\"400\">\r\n <img src=\"https://yeyupiaoling.cn/qq.png\" alt=\"QQ\u7fa4\" width=\"400\">\r\n</div>\r\n\r\n\u4f7f\u7528\u73af\u5883\uff1a\r\n\r\n - Anaconda 3\r\n - Python 3.11\r\n - Pytorch 2.4.0\r\n - Windows 11 or Ubuntu 22.04\r\n\r\n# \u76ee\u5f55\r\n\r\n- [\u9879\u76ee\u4ecb\u7ecd](#\u57fa\u4e8ePytorch\u5b9e\u73b0\u7684\u58f0\u7eb9\u8bc6\u522b\u7cfb\u7edf)\r\n- [\u9879\u76ee\u8bb0\u5f55](#\u9879\u76ee\u8bb0\u5f55)\r\n- [\u9879\u76ee\u7279\u6027](#\u9879\u76ee\u7279\u6027)\r\n- [\u5b89\u88c5\u73af\u5883](#\u5b89\u88c5\u73af\u5883)\r\n- [\u521b\u5efa\u6570\u636e](#\u521b\u5efa\u6570\u636e)\r\n- [\u4fee\u6539\u9884\u5904\u7406\u65b9\u6cd5\uff08\u53ef\u9009\uff09](#\u4fee\u6539\u9884\u5904\u7406\u65b9\u6cd5\u53ef\u9009)\r\n- [\u63d0\u53d6\u7279\u5f81\uff08\u53ef\u9009\uff09](#\u63d0\u53d6\u7279\u5f81\u53ef\u9009)\r\n- [\u8bad\u7ec3\u6a21\u578b](#\u8bad\u7ec3\u6a21\u578b)\r\n- [\u8bc4\u4f30\u6a21\u578b](#\u8bc4\u4f30\u6a21\u578b)\r\n- [\u63a8\u7406\u63a5\u53e3](#\u63a8\u7406\u63a5\u53e3)\r\n- [\u58f0\u7eb9\u5bf9\u6bd4](#\u58f0\u7eb9\u5bf9\u6bd4)\r\n- [\u58f0\u7eb9\u8bc6\u522b](#\u58f0\u7eb9\u8bc6\u522b)\r\n- [\u8bf4\u8bdd\u4eba\u65e5\u5fd7\uff08\u5206\u79bb\u8bf4\u8bdd\u4eba\uff09](#\u8bf4\u8bdd\u4eba\u65e5\u5fd7\u5206\u79bb\u8bf4\u8bdd\u4eba)\r\n\r\n\r\n# \u9879\u76ee\u8bb0\u5f55\r\n\r\n1. 2024.10.12\uff1a\u53d1\u5e031.1\u7248\u672c\u3002\r\n\r\n# \u9879\u76ee\u7279\u6027\r\n\r\n1. \u652f\u6301\u6a21\u578b\uff1aEcapaTdnn\u3001TDNN\u3001Res2Net\u3001ResNetSE\u3001ERes2Net\u3001CAM++\r\n2. \u652f\u6301\u6c60\u5316\u5c42\uff1aAttentiveStatsPool(ASP)\u3001SelfAttentivePooling(SAP)\u3001TemporalStatisticsPooling(TSP)\u3001TemporalAveragePooling(TAP)\u3001TemporalStatsPool(TSTP)\r\n3. \u652f\u6301\u635f\u5931\u51fd\u6570\uff1aAAMLoss\u3001SphereFace2\u3001AMLoss\u3001ARMLoss\u3001CELoss\u3001SubCenterLoss\u3001TripletAngularMarginLoss\r\n4. \u652f\u6301\u9884\u5904\u7406\u65b9\u6cd5\uff1aMelSpectrogram\u3001Spectrogram\u3001MFCC\u3001Fbank\u3001Wav2vec2.0\u3001WavLM\r\n5. \u652f\u6301\u6570\u636e\u589e\u5f3a\u65b9\u6cd5\uff1a\u8bed\u901f\u589e\u5f3a\u3001\u97f3\u91cf\u589e\u5f3a\u3001\u566a\u58f0\u589e\u5f3a\u3001\u6df7\u54cd\u589e\u5f3a\u3001SpecAugment\r\n\r\n\r\n**\u6a21\u578b\u8bba\u6587\uff1a**\r\n\r\n- EcapaTdnn\uff1a[ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification](https://arxiv.org/abs/2005.07143v3)\r\n- PANNS\uff1a[PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition](https://arxiv.org/abs/1912.10211v5)\r\n- TDNN\uff1a[Prediction of speech intelligibility with DNN-based performance measures](https://arxiv.org/abs/2203.09148)\r\n- Res2Net\uff1a[Res2Net: A New Multi-scale Backbone Architecture](https://arxiv.org/abs/1904.01169)\r\n- ResNetSE\uff1a[Squeeze-and-Excitation Networks](https://arxiv.org/abs/1709.01507)\r\n- CAMPPlus\uff1a[CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking](https://arxiv.org/abs/2303.00332v3)\r\n- ERes2Net\uff1a[An Enhanced Res2Net with Local and Global Feature Fusion for Speaker Verification](https://arxiv.org/abs/2305.12838v1)\r\n\r\n\r\n# \u6a21\u578b\u4e0b\u8f7d\r\n\r\n### \u8bad\u7ec3CN-Celeb\u6570\u636e\uff0c\u5171\u67092796\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n\r\n| \u6a21\u578b | Params(M) | \u6570\u636e\u96c6 | train speakers | threshold | EER | MinDCF | \u6a21\u578b\u4e0b\u8f7d |\r\n|:----------:|:---------:|:----------------------------------:|:--------------:|:---------:|:-------:|:-------:|:--------:|\r\n| ERes2NetV2 | 6.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.20089 | 0.08071 | 0.45705 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2Net | 6.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.20014 | 0.08132 | 0.45544 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CAM++ | 6.8 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23323 | 0.08332 | 0.48536 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ResNetSE | 7.8 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.19066 | 0.08544 | 0.49142 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| EcapaTdnn | 6.1 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23646 | 0.09259 | 0.51378 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| TDNN | 2.6 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.23858 | 0.10825 | 0.59545 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| Res2Net | 5.0 | [CN-Celeb](http://openslr.org/82/) | 2796 | 0.19526 | 0.12436 | 0.65347 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CAM++ | 6.8 | \u66f4\u5927\u6570\u636e\u96c6 | 2W+ | 0.33 | 0.07874 | 0.52524 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2Net | 55.1 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.36 | 0.02936 | 0.18355 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2NetV2 | 56.2 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.36 | 0.03847 | 0.24301 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CAM++ | 6.8 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.29 | 0.04765 | 0.31436 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n\r\n\u8bf4\u660e\uff1a\r\n1. \u8bc4\u4f30\u7684\u6d4b\u8bd5\u96c6\u4e3a[CN-Celeb\u7684\u6d4b\u8bd5\u96c6](https://aistudio.baidu.com/aistudio/datasetdetail/233361)\uff0c\u5305\u542b196\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n2. \u4f7f\u7528\u8bed\u901f\u589e\u5f3a\u5206\u7c7b\u5927\u5c0f\u7ffb\u4e09\u500d`speed_perturb_3_class: True`\u3002\r\n3. \u4f7f\u7528\u7684\u9884\u5904\u7406\u65b9\u6cd5\u4e3a`Fbank`\uff0c\u635f\u5931\u51fd\u6570\u4e3a`AAMLoss`\u3002\r\n4. \u53c2\u6570\u6570\u91cf\u4e0d\u5305\u542b\u4e86\u5206\u7c7b\u5668\u7684\u53c2\u6570\u6570\u91cf\u3002\r\n5. \u4f7f\u7528\u4e86\u566a\u58f0\u589e\u5f3a\u548c\u6df7\u54cd\u589e\u5f3a\u3002\r\n\r\n\r\n### \u8bad\u7ec3VoxCeleb1&2\u6570\u636e\uff0c\u5171\u67097205\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n\r\n| \u6a21\u578b | Params(M) | \u6570\u636e\u96c6 | train speakers | threshold | EER | MinDCF | \u6a21\u578b\u4e0b\u8f7d |\r\n|:----------:|:---------:|:-----------:|:--------------:|:---------:|:-------:|:-------:|:--------:|\r\n| CAM++ | 6.8 | VoxCeleb1&2 | 7205 | 0.22504 | 0.02436 | 0.15543 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| EcapaTdnn | 6.1 | VoxCeleb1&2 | 7205 | 0.24877 | 0.02480 | 0.16188 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ResNetSE | 7.8 | VoxCeleb1&2 | 7205 | 0.22567 | 0.03189 | 0.23040 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| TDNN | 2.6 | VoxCeleb1&2 | 7205 | 0.23834 | 0.03486 | 0.26792 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| Res2Net | 5.0 | VoxCeleb1&2 | 7205 | 0.19472 | 0.04370 | 0.40072 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2Net | 6.6 | VoxCeleb1&2 | 7205 | | | | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CAM++ | 6.8 | \u66f4\u5927\u6570\u636e\u96c6 | 2W+ | 0.28 | 0.03182 | 0.23731 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2Net | 55.1 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.53 | 0.08904 | 0.62130 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ERes2NetV2 | 56.2 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.52 | 0.08649 | 0.64193 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CAM++ | 6.8 | \u5176\u4ed6\u6570\u636e\u96c6 | 20W+ | 0.49 | 0.10334 | 0.71200 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n\r\n\u8bf4\u660e\uff1a\r\n\r\n1. \u8bc4\u4f30\u7684\u6d4b\u8bd5\u96c6\u4e3a[VoxCeleb1&2\u7684\u6d4b\u8bd5\u96c6](https://aistudio.baidu.com/aistudio/datasetdetail/255977)\uff0c\u5305\u542b158\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n2. \u4f7f\u7528\u8bed\u901f\u589e\u5f3a\u5206\u7c7b\u5927\u5c0f\u7ffb\u4e09\u500d`speed_perturb_3_class: True`\u3002\r\n3. \u4f7f\u7528\u7684\u9884\u5904\u7406\u65b9\u6cd5\u4e3a`Fbank`\uff0c\u635f\u5931\u51fd\u6570\u4e3a`AAMLoss`\u3002\r\n4. \u53c2\u6570\u6570\u91cf\u4e0d\u5305\u542b\u4e86\u5206\u7c7b\u5668\u7684\u53c2\u6570\u6570\u91cf\u3002\r\n\r\n\r\n### \u9884\u5904\u7406\u65b9\u6cd5\u6548\u679c\u5bf9\u6bd4\u5b9e\u9a8c\r\n\r\n| \u9884\u5904\u7406\u65b9\u6cd5 | \u6570\u636e\u96c6 | train speakers | threshold | EER | MinDCF | \u6a21\u578b\u4e0b\u8f7d |\r\n|:--------------------------------------------------------------------------------:|:--------:|:--------------:|:---------:|:-------:|:-------:|:--------:|\r\n| Fbank | CN-Celeb | 2796 | 0.14574 | 0.10988 | 0.58955 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| MFCC | CN-Celeb | 2796 | 0.14868 | 0.11483 | 0.61275 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| Spectrogram | CN-Celeb | 2796 | 0.14962 | 0.11613 | 0.60057 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| MelSpectrogram | CN-Celeb | 2796 | 0.13458 | 0.12498 | 0.60741 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| [wavlm-base-plus](https://huggingface.co/microsoft/wavlm-base-plus) | CN-Celeb | 2796 | 0.14166 | 0.13247 | 0.62451 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| [w2v-bert-2.0](https://huggingface.co/facebook/w2v-bert-2.0) | CN-Celeb | 2796 | | | | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| [wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) | CN-Celeb | 2796 | | | | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| [wavlm-large](https://huggingface.co/microsoft/wavlm-large) | CN-Celeb | 2796 | | | | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n\r\n\u8bf4\u660e\uff1a\r\n\r\n1. \u8bc4\u4f30\u7684\u6d4b\u8bd5\u96c6\u4e3a[CN-Celeb\u7684\u6d4b\u8bd5\u96c6](https://aistudio.baidu.com/aistudio/datasetdetail/233361)\uff0c\u5305\u542b196\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n2. \u5b9e\u9a8c\u6570\u636e\u4e3a[CN-Celeb](http://openslr.org/82/)\uff0c\u5b9e\u9a8c\u6a21\u578b\u4e3a`CAM++`\uff0c\u635f\u5931\u51fd\u6570\u4e3a`AAMLoss`\u3002\r\n3. \u6570\u636e\u4f7f\u7528`extract_features.py`\u63d0\u524d\u63d0\u53d6\u7279\u5f81\uff0c\u4e5f\u5c31\u662f\u8bf4\u8bad\u7ec3\u4e2d\u6ca1\u6709\u4f7f\u7528\u5bf9\u97f3\u9891\u7684\u6570\u636e\u589e\u5f3a\u3002\r\n4. `w2v-bert-2.0`\u3001`wav2vec2-large-xlsr-53`\u662f\u591a\u8bed\u8a00\u6570\u636e\u9884\u8bad\u7ec3\u5f97\u5230\u7684\uff0c`wavlm-base-plus`\u3001`wavlm-large`\u7684\u9884\u8bad\u7ec3\u6570\u636e\u4ec5\u7528\u82f1\u6587\u3002\r\n\r\n\r\n### \u635f\u5931\u51fd\u6570\u6548\u679c\u5bf9\u6bd4\u5b9e\u9a8c\r\n\r\n| \u635f\u5931\u51fd\u6570 | \u6570\u636e\u96c6 | train speakers | threshold | EER | MinDCF | \u6a21\u578b\u4e0b\u8f7d |\r\n|:------------------------:|:--------:|:--------------:|:---------:|:-------:|:-------:|:--------:|\r\n| AAMLoss | CN-Celeb | 2796 | 0.14574 | 0.10988 | 0.58955 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| SphereFace2 | CN-Celeb | 2796 | 0.20377 | 0.11309 | 0.61536 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| TripletAngularMarginLoss | CN-Celeb | 2796 | 0.28940 | 0.11749 | 0.63735 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| SubCenterLoss | CN-Celeb | 2796 | 0.13126 | 0.11775 | 0.56995 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| ARMLoss | CN-Celeb | 2796 | 0.14563 | 0.11805 | 0.57171 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| AMLoss | CN-Celeb | 2796 | 0.12870 | 0.12301 | 0.63263 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n| CELoss | CN-Celeb | 2796 | 0.13607 | 0.12684 | 0.65176 | \u52a0\u5165\u77e5\u8bc6\u661f\u7403\u83b7\u53d6 |\r\n\r\n\u8bf4\u660e\uff1a\r\n\r\n1. \u8bc4\u4f30\u7684\u6d4b\u8bd5\u96c6\u4e3a[CN-Celeb\u7684\u6d4b\u8bd5\u96c6](https://aistudio.baidu.com/aistudio/datasetdetail/233361)\uff0c\u5305\u542b196\u4e2a\u8bf4\u8bdd\u4eba\u3002\r\n2. \u5b9e\u9a8c\u6570\u636e\u4e3a[CN-Celeb](http://openslr.org/82/)\uff0c\u5b9e\u9a8c\u6a21\u578b\u4e3a`CAM++`\uff0c\u9884\u5904\u7406\u65b9\u6cd5\u4e3a`Fbank`\u3002\r\n3. \u6570\u636e\u4f7f\u7528`extract_features.py`\u63d0\u524d\u63d0\u53d6\u7279\u5f81\uff0c\u4e5f\u5c31\u662f\u8bf4\u8bad\u7ec3\u4e2d\u6ca1\u6709\u4f7f\u7528\u5bf9\u97f3\u9891\u7684\u6570\u636e\u589e\u5f3a\u3002\r\n\r\n\r\n## \u5b89\u88c5\u73af\u5883\r\n\r\n - \u9996\u5148\u5b89\u88c5\u7684\u662fPytorch\u7684GPU\u7248\u672c\uff0c\u5982\u679c\u5df2\u7ecf\u5b89\u88c5\u8fc7\u4e86\uff0c\u8bf7\u8df3\u8fc7\u3002\r\n```shell\r\nconda install pytorch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 pytorch-cuda=11.8 -c pytorch -c nvidia\r\n```\r\n\r\n - \u5b89\u88c5ppvector\u5e93\u3002\r\n \r\n\u4f7f\u7528pip\u5b89\u88c5\uff0c\u547d\u4ee4\u5982\u4e0b\uff1a\r\n```shell\r\npython -m pip install mvector -U -i https://pypi.tuna.tsinghua.edu.cn/simple\r\n```\r\n\r\n**\u5efa\u8bae\u6e90\u7801\u5b89\u88c5**\uff0c\u6e90\u7801\u5b89\u88c5\u80fd\u4fdd\u8bc1\u4f7f\u7528\u6700\u65b0\u4ee3\u7801\u3002\r\n```shell\r\ngit clone https://github.com/yeyupiaoling/VoiceprintRecognition-Pytorch.git\r\ncd VoiceprintRecognition-Pytorch/\r\npip install .\r\n```\r\n\r\n# \u521b\u5efa\u6570\u636e\r\n\u672c\u6559\u7a0b\u7b14\u8005\u4f7f\u7528\u7684\u662f[CN-Celeb](https://openslr.elda.org/resources/82)\uff0c\u8fd9\u4e2a\u6570\u636e\u96c6\u4e00\u5171\u6709\u7ea63000\u4e2a\u4eba\u7684\u8bed\u97f3\u6570\u636e\uff0c\u670965W+\u6761\u8bed\u97f3\u6570\u636e\uff0c\u4e0b\u8f7d\u4e4b\u540e\u8981\u89e3\u538b\u6570\u636e\u96c6\u5230`dataset`\u76ee\u5f55\uff0c\u53e6\u5916\u5982\u679c\u8981\u8bc4\u4f30\uff0c\u8fd8\u9700\u8981\u4e0b\u8f7d[CN-Celeb\u7684\u6d4b\u8bd5\u96c6](https://aistudio.baidu.com/aistudio/datasetdetail/233361)\u3002\u5982\u679c\u8bfb\u8005\u6709\u5176\u4ed6\u66f4\u597d\u7684\u6570\u636e\u96c6\uff0c\u53ef\u4ee5\u6df7\u5408\u5728\u4e00\u8d77\u4f7f\u7528\uff0c\u4f46\u6700\u597d\u662f\u8981\u7528python\u7684\u5de5\u5177\u6a21\u5757aukit\u5904\u7406\u97f3\u9891\uff0c\u964d\u566a\u548c\u53bb\u9664\u9759\u97f3\u3002\r\n\r\n\u9996\u5148\u662f\u521b\u5efa\u4e00\u4e2a\u6570\u636e\u5217\u8868\uff0c\u6570\u636e\u5217\u8868\u7684\u683c\u5f0f\u4e3a`<\u8bed\u97f3\u6587\u4ef6\u8def\u5f84\\t\u8bed\u97f3\u5206\u7c7b\u6807\u7b7e>`\uff0c\u521b\u5efa\u8fd9\u4e2a\u5217\u8868\u4e3b\u8981\u662f\u65b9\u4fbf\u4e4b\u540e\u7684\u8bfb\u53d6\uff0c\u4e5f\u662f\u65b9\u4fbf\u8bfb\u53d6\u4f7f\u7528\u5176\u4ed6\u7684\u8bed\u97f3\u6570\u636e\u96c6\uff0c\u8bed\u97f3\u5206\u7c7b\u6807\u7b7e\u662f\u6307\u8bf4\u8bdd\u4eba\u7684\u552f\u4e00ID\uff0c\u4e0d\u540c\u7684\u8bed\u97f3\u6570\u636e\u96c6\uff0c\u53ef\u4ee5\u901a\u8fc7\u7f16\u5199\u5bf9\u5e94\u7684\u751f\u6210\u6570\u636e\u5217\u8868\u7684\u51fd\u6570\uff0c\u628a\u8fd9\u4e9b\u6570\u636e\u96c6\u90fd\u5199\u5728\u540c\u4e00\u4e2a\u6570\u636e\u5217\u8868\u4e2d\u3002\r\n\r\n\u6267\u884c`create_data.py`\u7a0b\u5e8f\u5b8c\u6210\u6570\u636e\u51c6\u5907\u3002\r\n```shell\r\npython create_data.py\r\n```\r\n\r\n\u6267\u884c\u4e0a\u9762\u7684\u7a0b\u5e8f\u4e4b\u540e\uff0c\u4f1a\u751f\u6210\u4ee5\u4e0b\u7684\u6570\u636e\u683c\u5f0f\uff0c\u5982\u679c\u8981\u81ea\u5b9a\u4e49\u6570\u636e\uff0c\u53c2\u8003\u5982\u4e0b\u6570\u636e\u5217\u8868\uff0c\u524d\u9762\u662f\u97f3\u9891\u7684\u76f8\u5bf9\u8def\u5f84\uff0c\u540e\u9762\u7684\u662f\u8be5\u97f3\u9891\u5bf9\u5e94\u7684\u8bf4\u8bdd\u4eba\u7684\u6807\u7b7e\uff0c\u5c31\u8ddf\u5206\u7c7b\u4e00\u6837\u3002**\u81ea\u5b9a\u4e49\u6570\u636e\u96c6\u7684\u6ce8\u610f**\uff0c\u6d4b\u8bd5\u6570\u636e\u5217\u8868\u7684ID\u53ef\u4ee5\u4e0d\u7528\u8ddf\u8bad\u7ec3\u7684ID\u4e00\u6837\uff0c\u4e5f\u5c31\u662f\u8bf4\u6d4b\u8bd5\u7684\u6570\u636e\u7684\u8bf4\u8bdd\u4eba\u53ef\u4ee5\u4e0d\u7528\u51fa\u73b0\u5728\u8bad\u7ec3\u96c6\uff0c\u53ea\u8981\u4fdd\u8bc1\u6d4b\u8bd5\u6570\u636e\u5217\u8868\u4e2d\u540c\u4e00\u4e2a\u4eba\u76f8\u540c\u7684ID\u5373\u53ef\u3002\r\n```\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-03-019.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-10-023.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-06-025.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-04-014.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-06-030.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-10-032.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-06-028.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-10-031.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-05-003.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-04-017.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-10-016.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-09-001.flac 2795\r\ndataset/CN-Celeb2_flac/data/id11999/recitation-05-010.flac 2795\r\n```\r\n\r\n# \u4fee\u6539\u9884\u5904\u7406\u65b9\u6cd5\uff08\u53ef\u9009\uff09\r\n\r\n\u914d\u7f6e\u6587\u4ef6\u4e2d\u9ed8\u8ba4\u4f7f\u7528\u7684\u662fFbank\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u5982\u679c\u8981\u4f7f\u7528\u5176\u4ed6\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u53ef\u4ee5\u4fee\u6539\u914d\u7f6e\u6587\u4ef6\u4e2d\u7684\u5b89\u88c5\u4e0b\u9762\u65b9\u5f0f\u4fee\u6539\uff0c\u5177\u4f53\u7684\u503c\u53ef\u4ee5\u6839\u636e\u81ea\u5df1\u60c5\u51b5\u4fee\u6539\u3002\u5982\u679c\u4e0d\u6e05\u695a\u5982\u4f55\u8bbe\u7f6e\u53c2\u6570\uff0c\u53ef\u4ee5\u76f4\u63a5\u5220\u9664\u8be5\u90e8\u5206\uff0c\u76f4\u63a5\u4f7f\u7528\u9ed8\u8ba4\u503c\u3002\r\n\r\n```yaml\r\n# \u6570\u636e\u9884\u5904\u7406\u53c2\u6570\r\npreprocess_conf:\r\n # \u662f\u5426\u4f7f\u7528HF\u4e0a\u7684Wav2Vec2\u7c7b\u4f3c\u6a21\u578b\u63d0\u53d6\u97f3\u9891\u7279\u5f81\r\n use_hf_model: False\r\n # \u97f3\u9891\u9884\u5904\u7406\u65b9\u6cd5\uff0c\u4e5f\u53ef\u4ee5\u53eb\u7279\u5f81\u63d0\u53d6\u65b9\u6cd5\r\n # \u5f53use_hf_model\u4e3aFalse\u65f6\uff0c\u652f\u6301\uff1aMelSpectrogram\u3001Spectrogram\u3001MFCC\u3001Fbank\r\n # \u5f53use_hf_model\u4e3aTrue\u65f6\uff0c\u6307\u5b9a\u7684\u662fHuggingFace\u7684\u6a21\u578b\u6216\u8005\u672c\u5730\u8def\u5f84\uff0c\u6bd4\u5982facebook/w2v-bert-2.0\u6216\u8005./feature_models/w2v-bert-2.0\r\n feature_method: 'Fbank'\r\n # \u5f53use_hf_model\u4e3aFalse\u65f6\uff0c\u8bbe\u7f6eAPI\u53c2\u6570\uff0c\u66f4\u53c2\u6570\u67e5\u770b\u5bf9\u5e94API\uff0c\u4e0d\u6e05\u695a\u7684\u53ef\u4ee5\u76f4\u63a5\u5220\u9664\u8be5\u90e8\u5206\uff0c\u76f4\u63a5\u4f7f\u7528\u9ed8\u8ba4\u503c\u3002\r\n # \u5f53use_hf_model\u4e3aTrue\u65f6\uff0c\u53ef\u4ee5\u8bbe\u7f6e\u53c2\u6570use_gpu\uff0c\u6307\u5b9a\u662f\u5426\u4f7f\u7528GPU\u63d0\u53d6\u7279\u5f81\r\n method_args:\r\n sample_frequency: 16000\r\n num_mel_bins: 80\r\n```\r\n\r\n# \u63d0\u53d6\u7279\u5f81\uff08\u53ef\u9009\uff09\r\n\r\n\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u9996\u5148\u662f\u8981\u8bfb\u53d6\u97f3\u9891\u6570\u636e\uff0c\u7136\u540e\u63d0\u53d6\u7279\u5f81\uff0c\u6700\u540e\u518d\u8fdb\u884c\u8bad\u7ec3\u3002\u5176\u4e2d\u8bfb\u53d6\u97f3\u9891\u6570\u636e\u3001\u63d0\u53d6\u7279\u5f81\u4e5f\u662f\u6bd4\u8f83\u6d88\u8017\u65f6\u95f4\u7684\uff0c\u6240\u4ee5\u6211\u4eec\u53ef\u4ee5\u9009\u62e9\u63d0\u524d\u63d0\u53d6\u597d\u53d6\u7279\u5f81\uff0c\u8bad\u7ec3\u6a21\u578b\u7684\u662f\u5c31\u53ef\u4ee5\u76f4\u63a5\u52a0\u8f7d\u63d0\u53d6\u597d\u7684\u7279\u5f81\uff0c\u8fd9\u6837\u8bad\u7ec3\u901f\u5ea6\u4f1a\u66f4\u5feb\u3002\u8fd9\u4e2a\u63d0\u53d6\u7279\u5f81\u662f\u53ef\u9009\u62e9\uff0c\u5982\u679c\u6ca1\u6709\u63d0\u53d6\u597d\u7684\u7279\u5f81\uff0c\u8bad\u7ec3\u6a21\u578b\u7684\u65f6\u5019\u5c31\u4f1a\u4ece\u8bfb\u53d6\u97f3\u9891\u6570\u636e\uff0c\u7136\u540e\u63d0\u53d6\u7279\u5f81\u5f00\u59cb\u3002\u63d0\u53d6\u7279\u5f81\u6b65\u9aa4\u5982\u4e0b\uff1a\r\n\r\n1. \u6267\u884c`extract_features.py`\uff0c\u63d0\u53d6\u7279\u5f81\uff0c\u7279\u5f81\u4f1a\u4fdd\u5b58\u5728`dataset/features`\u76ee\u5f55\u4e0b\uff0c\u5e76\u751f\u6210\u65b0\u7684\u6570\u636e\u5217\u8868`train_list_features.txt`\u3001`enroll_list_features.txt`\u548c`trials_list_features.txt`\u3002\r\n\r\n```shell\r\npython extract_features.py --configs=configs/cam++.yml --save_dir=dataset/features\r\n```\r\n\r\n2. \u4fee\u6539\u914d\u7f6e\u6587\u4ef6\uff0c\u5c06`dataset_conf.train_list`\u3001`dataset_conf.enroll_list`\u548c`dataset_conf.trials_list`\u4fee\u6539\u4e3a`train_list_features.txt`\u3001`enroll_list_features.txt`\u548c`trials_list_features.txt`\u3002\r\n\r\n\r\n# \u8bad\u7ec3\u6a21\u578b\r\n\u4f7f\u7528`train.py`\u8bad\u7ec3\u6a21\u578b\uff0c\u672c\u9879\u76ee\u652f\u6301\u591a\u4e2a\u97f3\u9891\u9884\u5904\u7406\u65b9\u5f0f\uff0c\u901a\u8fc7`configs/ecapa_tdnn.yml`\u914d\u7f6e\u6587\u4ef6\u7684\u53c2\u6570`preprocess_conf.feature_method`\u53ef\u4ee5\u6307\u5b9a\uff0c`MelSpectrogram`\u4e3a\u6885\u5c14\u9891\u8c31\uff0c`Spectrogram`\u4e3a\u8bed\u8c31\u56fe\uff0c`MFCC`\u6885\u5c14\u9891\u8c31\u5012\u8c31\u7cfb\u6570\u7b49\u7b49\u3002\u901a\u8fc7\u53c2\u6570`augment_conf_path`\u53ef\u4ee5\u6307\u5b9a\u6570\u636e\u589e\u5f3a\u65b9\u5f0f\u3002\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u4f1a\u4f7f\u7528VisualDL\u4fdd\u5b58\u8bad\u7ec3\u65e5\u5fd7\uff0c\u901a\u8fc7\u542f\u52a8VisualDL\u53ef\u4ee5\u968f\u65f6\u67e5\u770b\u8bad\u7ec3\u7ed3\u679c\uff0c\u542f\u52a8\u547d\u4ee4`visualdl --logdir=log --host 0.0.0.0`\r\n```shell\r\n# \u5355\u5361\u8bad\u7ec3\r\nCUDA_VISIBLE_DEVICES=0 python train.py\r\n# \u591a\u5361\u8bad\u7ec3\r\nCUDA_VISIBLE_DEVICES=0,1 torchrun --standalone --nnodes=1 --nproc_per_node=2 train.py\r\n```\r\n\r\n\u8bad\u7ec3\u8f93\u51fa\u65e5\u5fd7\uff1a\r\n```\r\n[2023-08-05 09:52:06.497988 INFO ] utils:print_arguments:13 - ----------- \u989d\u5916\u914d\u7f6e\u53c2\u6570 -----------\r\n[2023-08-05 09:52:06.498094 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml\r\n[2023-08-05 09:52:06.498149 INFO ] utils:print_arguments:15 - do_eval: True\r\n[2023-08-05 09:52:06.498191 INFO ] utils:print_arguments:15 - local_rank: 0\r\n[2023-08-05 09:52:06.498230 INFO ] utils:print_arguments:15 - pretrained_model: None\r\n[2023-08-05 09:52:06.498269 INFO ] utils:print_arguments:15 - resume_model: None\r\n[2023-08-05 09:52:06.498306 INFO ] utils:print_arguments:15 - save_model_path: models/\r\n[2023-08-05 09:52:06.498342 INFO ] utils:print_arguments:15 - use_gpu: True\r\n[2023-08-05 09:52:06.498378 INFO ] utils:print_arguments:16 - ------------------------------------------------\r\n[2023-08-05 09:52:06.513761 INFO ] utils:print_arguments:18 - ----------- \u914d\u7f6e\u6587\u4ef6\u53c2\u6570 -----------\r\n[2023-08-05 09:52:06.513906 INFO ] utils:print_arguments:21 - dataset_conf:\r\n[2023-08-05 09:52:06.513957 INFO ] utils:print_arguments:24 - dataLoader:\r\n[2023-08-05 09:52:06.513995 INFO ] utils:print_arguments:26 - batch_size: 64\r\n[2023-08-05 09:52:06.514031 INFO ] utils:print_arguments:26 - num_workers: 4\r\n[2023-08-05 09:52:06.514066 INFO ] utils:print_arguments:28 - do_vad: False\r\n[2023-08-05 09:52:06.514101 INFO ] utils:print_arguments:28 - enroll_list: dataset/enroll_list.txt\r\n[2023-08-05 09:52:06.514135 INFO ] utils:print_arguments:24 - eval_conf:\r\n[2023-08-05 09:52:06.514169 INFO ] utils:print_arguments:26 - batch_size: 1\r\n[2023-08-05 09:52:06.514203 INFO ] utils:print_arguments:26 - max_duration: 20\r\n[2023-08-05 09:52:06.514237 INFO ] utils:print_arguments:28 - max_duration: 3\r\n[2023-08-05 09:52:06.514274 INFO ] utils:print_arguments:28 - min_duration: 0.5\r\n[2023-08-05 09:52:06.514308 INFO ] utils:print_arguments:28 - noise_aug_prob: 0.2\r\n[2023-08-05 09:52:06.514342 INFO ] utils:print_arguments:28 - noise_dir: dataset/noise\r\n[2023-08-05 09:52:06.514374 INFO ] utils:print_arguments:28 - num_speakers: 3242\r\n[2023-08-05 09:52:06.514408 INFO ] utils:print_arguments:28 - sample_rate: 16000\r\n[2023-08-05 09:52:06.514441 INFO ] utils:print_arguments:28 - speed_perturb: True\r\n[2023-08-05 09:52:06.514475 INFO ] utils:print_arguments:28 - target_dB: -20\r\n[2023-08-05 09:52:06.514508 INFO ] utils:print_arguments:28 - train_list: dataset/train_list.txt\r\n[2023-08-05 09:52:06.514542 INFO ] utils:print_arguments:28 - trials_list: dataset/trials_list.txt\r\n[2023-08-05 09:52:06.514575 INFO ] utils:print_arguments:28 - use_dB_normalization: True\r\n[2023-08-05 09:52:06.514609 INFO ] utils:print_arguments:21 - loss_conf:\r\n[2023-08-05 09:52:06.514643 INFO ] utils:print_arguments:24 - args:\r\n[2023-08-05 09:52:06.514678 INFO ] utils:print_arguments:26 - easy_margin: False\r\n[2023-08-05 09:52:06.514713 INFO ] utils:print_arguments:26 - margin: 0.2\r\n[2023-08-05 09:52:06.514746 INFO ] utils:print_arguments:26 - scale: 32\r\n[2023-08-05 09:52:06.514779 INFO ] utils:print_arguments:24 - margin_scheduler_args:\r\n[2023-08-05 09:52:06.514814 INFO ] utils:print_arguments:26 - final_margin: 0.3\r\n[2023-08-05 09:52:06.514848 INFO ] utils:print_arguments:28 - use_loss: AAMLoss\r\n[2023-08-05 09:52:06.514882 INFO ] utils:print_arguments:28 - use_margin_scheduler: True\r\n[2023-08-05 09:52:06.514915 INFO ] utils:print_arguments:21 - model_conf:\r\n[2023-08-05 09:52:06.514950 INFO ] utils:print_arguments:24 - backbone:\r\n[2023-08-05 09:52:06.514984 INFO ] utils:print_arguments:26 - embd_dim: 192\r\n[2023-08-05 09:52:06.515017 INFO ] utils:print_arguments:26 - pooling_type: ASP\r\n[2023-08-05 09:52:06.515050 INFO ] utils:print_arguments:24 - classifier:\r\n[2023-08-05 09:52:06.515084 INFO ] utils:print_arguments:26 - num_blocks: 0\r\n[2023-08-05 09:52:06.515118 INFO ] utils:print_arguments:21 - optimizer_conf:\r\n[2023-08-05 09:52:06.515154 INFO ] utils:print_arguments:28 - learning_rate: 0.001\r\n[2023-08-05 09:52:06.515188 INFO ] utils:print_arguments:28 - optimizer: Adam\r\n[2023-08-05 09:52:06.515221 INFO ] utils:print_arguments:28 - scheduler: CosineAnnealingLR\r\n[2023-08-05 09:52:06.515254 INFO ] utils:print_arguments:28 - scheduler_args: None\r\n[2023-08-05 09:52:06.515289 INFO ] utils:print_arguments:28 - weight_decay: 1e-06\r\n[2023-08-05 09:52:06.515323 INFO ] utils:print_arguments:21 - preprocess_conf:\r\n[2023-08-05 09:52:06.515357 INFO ] utils:print_arguments:28 - feature_method: MelSpectrogram\r\n[2023-08-05 09:52:06.515390 INFO ] utils:print_arguments:24 - method_args:\r\n[2023-08-05 09:52:06.515426 INFO ] utils:print_arguments:26 - f_max: 14000.0\r\n[2023-08-05 09:52:06.515460 INFO ] utils:print_arguments:26 - f_min: 50.0\r\n[2023-08-05 09:52:06.515493 INFO ] utils:print_arguments:26 - hop_length: 320\r\n[2023-08-05 09:52:06.515527 INFO ] utils:print_arguments:26 - n_fft: 1024\r\n[2023-08-05 09:52:06.515560 INFO ] utils:print_arguments:26 - n_mels: 64\r\n[2023-08-05 09:52:06.515593 INFO ] utils:print_arguments:26 - sample_rate: 16000\r\n[2023-08-05 09:52:06.515626 INFO ] utils:print_arguments:26 - win_length: 1024\r\n[2023-08-05 09:52:06.515660 INFO ] utils:print_arguments:21 - train_conf:\r\n[2023-08-05 09:52:06.515694 INFO ] utils:print_arguments:28 - log_interval: 100\r\n[2023-08-05 09:52:06.515728 INFO ] utils:print_arguments:28 - max_epoch: 30\r\n[2023-08-05 09:52:06.515761 INFO ] utils:print_arguments:30 - use_model: EcapaTdnn\r\n[2023-08-05 09:52:06.515794 INFO ] utils:print_arguments:31 - ------------------------------------------------\r\n\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\r\n===============================================================================================\r\nLayer (type:depth-idx) Output Shape Param #\r\n===============================================================================================\r\nSequential [1, 9726] --\r\n\u251c\u2500EcapaTdnn: 1-1 [1, 192] --\r\n\u2502 \u2514\u2500Conv1dReluBn: 2-1 [1, 512, 98] --\r\n\u2502 \u2502 \u2514\u2500Conv1d: 3-1 [1, 512, 98] 163,840\r\n\u2502 \u2502 \u2514\u2500BatchNorm1d: 3-2 [1, 512, 98] 1,024\r\n\u2502 \u2514\u2500Sequential: 2-2 [1, 512, 98] --\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-3 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500Res2Conv1dReluBn: 3-4 [1, 512, 98] 86,912\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-5 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500SE_Connect: 3-6 [1, 512, 98] 262,912\r\n\u2502 \u2514\u2500Sequential: 2-3 [1, 512, 98] --\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-7 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500Res2Conv1dReluBn: 3-8 [1, 512, 98] 86,912\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-9 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500SE_Connect: 3-10 [1, 512, 98] 262,912\r\n\u2502 \u2514\u2500Sequential: 2-4 [1, 512, 98] --\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-11 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500Res2Conv1dReluBn: 3-12 [1, 512, 98] 86,912\r\n\u2502 \u2502 \u2514\u2500Conv1dReluBn: 3-13 [1, 512, 98] 263,168\r\n\u2502 \u2502 \u2514\u2500SE_Connect: 3-14 [1, 512, 98] 262,912\r\n\u2502 \u2514\u2500Conv1d: 2-5 [1, 1536, 98] 2,360,832\r\n\u2502 \u2514\u2500AttentiveStatsPool: 2-6 [1, 3072] --\r\n\u2502 \u2502 \u2514\u2500Conv1d: 3-15 [1, 128, 98] 196,736\r\n\u2502 \u2502 \u2514\u2500Conv1d: 3-16 [1, 1536, 98] 198,144\r\n\u2502 \u2514\u2500BatchNorm1d: 2-7 [1, 3072] 6,144\r\n\u2502 \u2514\u2500Linear: 2-8 [1, 192] 590,016\r\n\u2502 \u2514\u2500BatchNorm1d: 2-9 [1, 192] 384\r\n\u251c\u2500SpeakerIdentification: 1-2 [1, 9726] 1,867,392\r\n===============================================================================================\r\nTotal params: 8,012,992\r\nTrainable params: 8,012,992\r\nNon-trainable params: 0\r\nTotal mult-adds (M): 468.81\r\n===============================================================================================\r\nInput size (MB): 0.03\r\nForward/backward pass size (MB): 10.36\r\nParams size (MB): 32.05\r\nEstimated Total Size (MB): 42.44\r\n===============================================================================================\r\n[2023-08-05 09:52:08.084231 INFO ] trainer:train:388 - \u8bad\u7ec3\u6570\u636e\uff1a874175\r\n[2023-08-05 09:52:09.186542 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [0/13659], loss: 11.95824, accuracy: 0.00000, learning rate: 0.00100000, speed: 58.09 data/sec, eta: 5 days, 5:24:08\r\n[2023-08-05 09:52:22.477905 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [100/13659], loss: 10.35675, accuracy: 0.00278, learning rate: 0.00100000, speed: 481.65 data/sec, eta: 15:07:15\r\n[2023-08-05 09:52:35.948581 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [200/13659], loss: 10.22089, accuracy: 0.00505, learning rate: 0.00100000, speed: 475.27 data/sec, eta: 15:19:12\r\n[2023-08-05 09:52:49.249098 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [300/13659], loss: 10.00268, accuracy: 0.00706, learning rate: 0.00100000, speed: 481.45 data/sec, eta: 15:07:11\r\n[2023-08-05 09:53:03.716015 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [400/13659], loss: 9.76052, accuracy: 0.00830, learning rate: 0.00100000, speed: 442.74 data/sec, eta: 16:26:16\r\n[2023-08-05 09:53:18.258807 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [500/13659], loss: 9.50189, accuracy: 0.01060, learning rate: 0.00100000, speed: 440.46 data/sec, eta: 16:31:08\r\n[2023-08-05 09:53:31.618354 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [600/13659], loss: 9.26083, accuracy: 0.01256, learning rate: 0.00100000, speed: 479.50 data/sec, eta: 15:10:12\r\n[2023-08-05 09:53:45.439642 INFO ] trainer:__train_epoch:334 - Train epoch: [1/30], batch: [700/13659], loss: 9.03548, accuracy: 0.01449, learning rate: 0.00099999, speed: 463.63 data/sec, eta: 15:41:08\r\n```\r\n\r\n\u542f\u52a8VisualDL\uff1a`visualdl --logdir=log --host 0.0.0.0`\uff0cVisualDL\u9875\u9762\u5982\u4e0b\uff1a\r\n\r\n<div align=\"center\">\r\n<img src=\"./docs/images/log.jpg\" alt=\"VisualDL\u9875\u9762\" width=\"600\">\r\n</div>\r\n\r\n\r\n# \u8bc4\u4f30\u6a21\u578b\r\n\u8bad\u7ec3\u7ed3\u675f\u4e4b\u540e\u4f1a\u4fdd\u5b58\u9884\u6d4b\u6a21\u578b\uff0c\u6211\u4eec\u7528\u9884\u6d4b\u6a21\u578b\u6765\u9884\u6d4b\u6d4b\u8bd5\u96c6\u4e2d\u7684\u97f3\u9891\u7279\u5f81\uff0c\u7136\u540e\u4f7f\u7528\u97f3\u9891\u7279\u5f81\u8fdb\u884c\u4e24\u4e24\u5bf9\u6bd4\uff0c\u8ba1\u7b97EER\u548cMinDCF\u3002\r\n```shell\r\npython eval.py\r\n```\r\n\r\n\u8f93\u51fa\u7c7b\u4f3c\u5982\u4e0b\uff1a\r\n```\r\n\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\r\n------------------------------------------------\r\nW0425 08:27:32.057426 17654 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2\r\nW0425 08:27:32.065165 17654 device_context.cc:465] device: 0, cuDNN Version: 7.6.\r\n[2023-03-16 20:20:47.195908 INFO ] trainer:evaluate:341 - \u6210\u529f\u52a0\u8f7d\u6a21\u578b\uff1amodels/EcapaTdnn_Fbank/best_model/model.pth\r\n100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 84/84 [00:28<00:00, 2.95it/s]\r\n\u5f00\u59cb\u4e24\u4e24\u5bf9\u6bd4\u97f3\u9891\u7279\u5f81...\r\n100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 5332/5332 [00:05<00:00, 1027.83it/s]\r\n\u8bc4\u4f30\u6d88\u8017\u65f6\u95f4\uff1a65s\uff0cthreshold\uff1a0.26\uff0cEER: 0.14739, MinDCF: 0.41999\r\n```\r\n\r\n# \u63a8\u7406\u63a5\u53e3\r\n\r\n\u4e0b\u9762\u7ed9\u51fa\u4e86\u51e0\u4e2a\u5e38\u7528\u7684\u63a5\u53e3\uff0c\u66f4\u591a\u63a5\u53e3\u8bf7\u53c2\u8003`mvector/predict.py`\uff0c\u4e5f\u53ef\u4ee5\u5f80\u4e0b\u770b`\u58f0\u7eb9\u5bf9\u6bd4`\u548c`\u58f0\u7eb9\u8bc6\u522b`\u7684\u4f8b\u5b50\u3002\r\n\r\n```python\r\nfrom mvector.predict import MVectorPredictor\r\n\r\npredictor = MVectorPredictor(configs='configs/cam++.yml',\r\n model_path='models/CAMPPlus_Fbank/best_model/')\r\n# \u83b7\u53d6\u97f3\u9891\u7279\u5f81\r\nembedding = predictor.predict(audio_data='dataset/a_1.wav')\r\n# \u83b7\u53d6\u4e24\u4e2a\u97f3\u9891\u7684\u76f8\u4f3c\u5ea6\r\nsimilarity = predictor.contrast(audio_data1='dataset/a_1.wav', audio_data2='dataset/a_2.wav')\r\n\r\n# \u6ce8\u518c\u7528\u6237\u97f3\u9891\r\npredictor.register(user_name='\u591c\u96e8\u98d8\u96f6', audio_data='dataset/test.wav')\r\n# \u8bc6\u522b\u7528\u6237\u97f3\u9891\r\nname, score = predictor.recognition(audio_data='dataset/test1.wav')\r\n# \u83b7\u53d6\u6240\u6709\u7528\u6237\r\nusers_name = predictor.get_users()\r\n# \u5220\u9664\u7528\u6237\u97f3\u9891\r\npredictor.remove_user(user_name='\u591c\u96e8\u98d8\u96f6')\r\n```\r\n\r\n# \u58f0\u7eb9\u5bf9\u6bd4\r\n\u4e0b\u9762\u5f00\u59cb\u5b9e\u73b0\u58f0\u7eb9\u5bf9\u6bd4\uff0c\u521b\u5efa`infer_contrast.py`\u7a0b\u5e8f\uff0c\u9996\u5148\u4ecb\u7ecd\u51e0\u4e2a\u91cd\u8981\u7684\u51fd\u6570\uff0c`predict()`\u51fd\u6570\u662f\u53ef\u4ee5\u83b7\u53d6\u58f0\u7eb9\u7279\u5f81\uff0c`predict_batch()`\u51fd\u6570\u662f\u53ef\u4ee5\u83b7\u53d6\u4e00\u6279\u7684\u58f0\u7eb9\u7279\u5f81\uff0c`contrast()`\u51fd\u6570\u53ef\u4ee5\u5bf9\u6bd4\u4e24\u6761\u97f3\u9891\u7684\u76f8\u4f3c\u5ea6\uff0c`register()`\u51fd\u6570\u6ce8\u518c\u4e00\u6761\u97f3\u9891\u5230\u58f0\u7eb9\u5e93\u91cc\u9762\uff0c`recognition()`\u51fd\u8f93\u5165\u4e00\u6761\u97f3\u9891\u5e76\u4e14\u4ece\u58f0\u7eb9\u5e93\u91cc\u9762\u5bf9\u6bd4\u8bc6\u522b\uff0c`remove_user()`\u51fd\u6570\u79fb\u9664\u4f60\u597d\u3002\u58f0\u7eb9\u5e93\u91cc\u9762\u7684\u6ce8\u518c\u4eba\u3002\u6211\u4eec\u8f93\u5165\u4e24\u4e2a\u8bed\u97f3\uff0c\u901a\u8fc7\u9884\u6d4b\u51fd\u6570\u83b7\u53d6\u4ed6\u4eec\u7684\u7279\u5f81\u6570\u636e\uff0c\u4f7f\u7528\u8fd9\u4e2a\u7279\u5f81\u6570\u636e\u53ef\u4ee5\u6c42\u4ed6\u4eec\u7684\u5bf9\u89d2\u4f59\u5f26\u503c\uff0c\u5f97\u5230\u7684\u7ed3\u679c\u53ef\u4ee5\u4f5c\u4e3a\u4ed6\u4eec\u76f8\u8bc6\u5ea6\u3002\u5bf9\u4e8e\u8fd9\u4e2a\u76f8\u8bc6\u5ea6\u7684\u9608\u503c`threshold`\uff0c\u8bfb\u8005\u53ef\u4ee5\u6839\u636e\u81ea\u5df1\u9879\u76ee\u7684\u51c6\u786e\u5ea6\u8981\u6c42\u8fdb\u884c\u4fee\u6539\u3002\r\n```shell\r\npython infer_contrast.py --audio_path1=audio/a_1.wav --audio_path2=audio/b_2.wav\r\n```\r\n\r\n\u8f93\u51fa\u7c7b\u4f3c\u5982\u4e0b\uff1a\r\n```\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:13 - ----------- \u989d\u5916\u914d\u7f6e\u53c2\u6570 -----------\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - audio_path1: dataset/a_1.wav\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - audio_path2: dataset/b_2.wav\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - model_path: models/EcapaTdnn_Fbank/best_model/\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - threshold: 0.6\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:15 - use_gpu: True\r\n[2023-04-02 18:30:48.009149 INFO ] utils:print_arguments:16 - ------------------------------------------------\r\n\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\r\nW0425 08:29:10.006249 21121 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2\r\nW0425 08:29:10.008555 21121 device_context.cc:465] device: 0, cuDNN Version: 7.6.\r\n\u6210\u529f\u52a0\u8f7d\u6a21\u578b\u53c2\u6570\u548c\u4f18\u5316\u65b9\u6cd5\u53c2\u6570\uff1amodels/EcapaTdnn_Fbank/best_model/model.pth\r\naudio/a_1.wav \u548c audio/b_2.wav \u4e0d\u662f\u540c\u4e00\u4e2a\u4eba\uff0c\u76f8\u4f3c\u5ea6\u4e3a\uff1a-0.09565544128417969\r\n```\r\n\r\n\u540c\u65f6\u8fd8\u63d0\u4f9b\u4e86\u6709GUI\u754c\u9762\u7684\u58f0\u7eb9\u5bf9\u6bd4\u7a0b\u5e8f\uff0c\u6267\u884c`infer_contrast_gui.py`\u542f\u52a8\u7a0b\u5e8f\uff0c\u754c\u9762\u5982\u4e0b\uff0c\u5206\u522b\u9009\u62e9\u4e24\u4e2a\u97f3\u9891\uff0c\u70b9\u51fb\u5f00\u59cb\u5224\u65ad\uff0c\u5c31\u53ef\u4ee5\u5224\u65ad\u5b83\u4eec\u662f\u5426\u662f\u540c\u4e00\u4e2a\u4eba\u3002\r\n\r\n<div align=\"center\">\r\n<img src=\"./docs/images/contrast.jpg\" alt=\"\u58f0\u7eb9\u5bf9\u6bd4\u754c\u9762\">\r\n</div>\r\n\r\n# \u58f0\u7eb9\u8bc6\u522b\r\n\r\n\u5728\u65b0\u95fb\u8bc6\u522b\u91cc\u9762\u4e3b\u8981\u4f7f\u7528\u5230`register()`\u51fd\u6570\u548c`recognition()`\u51fd\u6570\uff0c\u9996\u5148\u4f7f\u7528`register()`\u51fd\u6570\u51fd\u6570\u6765\u6ce8\u518c\u97f3\u9891\u5230\u58f0\u7eb9\u5e93\u91cc\u9762\uff0c\u4e5f\u53ef\u4ee5\u76f4\u63a5\u628a\u6587\u4ef6\u6dfb\u52a0\u5230`audio_db`\u6587\u4ef6\u5939\u91cc\u9762\uff0c\u4f7f\u7528\u7684\u65f6\u5019\u901a\u8fc7`recognition()`\u51fd\u6570\u6765\u53d1\u8d77\u8bc6\u522b\uff0c\u8f93\u5165\u4e00\u6761\u97f3\u9891\uff0c\u5c31\u53ef\u4ee5\u4ece\u58f0\u7eb9\u5e93\u91cc\u9762\u8bc6\u522b\u5230\u6240\u9700\u8981\u7684\u8bf4\u8bdd\u4eba\u3002\r\n\r\n\u6709\u4e86\u4e0a\u9762\u7684\u58f0\u7eb9\u8bc6\u522b\u7684\u51fd\u6570\uff0c\u8bfb\u8005\u53ef\u4ee5\u6839\u636e\u81ea\u5df1\u9879\u76ee\u7684\u9700\u6c42\u5b8c\u6210\u58f0\u7eb9\u8bc6\u522b\u7684\u65b9\u5f0f\uff0c\u4f8b\u5982\u7b14\u8005\u4e0b\u9762\u63d0\u4f9b\u7684\u662f\u901a\u8fc7\u5f55\u97f3\u6765\u5b8c\u6210\u58f0\u7eb9\u8bc6\u522b\u3002\u9996\u5148\u5fc5\u987b\u8981\u52a0\u8f7d\u8bed\u97f3\u5e93\u4e2d\u7684\u8bed\u97f3\uff0c\u8bed\u97f3\u5e93\u6587\u4ef6\u5939\u4e3a`audio_db`\uff0c\u7136\u540e\u7528\u6237\u56de\u8f66\u540e\u5f55\u97f33\u79d2\u949f\uff0c\u7136\u540e\u7a0b\u5e8f\u4f1a\u81ea\u52a8\u5f55\u97f3\uff0c\u5e76\u4f7f\u7528\u5f55\u97f3\u5230\u7684\u97f3\u9891\u8fdb\u884c\u58f0\u7eb9\u8bc6\u522b\uff0c\u53bb\u5339\u914d\u8bed\u97f3\u5e93\u4e2d\u7684\u8bed\u97f3\uff0c\u83b7\u53d6\u7528\u6237\u7684\u4fe1\u606f\u3002\u901a\u8fc7\u8fd9\u6837\u65b9\u5f0f\uff0c\u8bfb\u8005\u4e5f\u53ef\u4ee5\u4fee\u6539\u6210\u901a\u8fc7\u670d\u52a1\u8bf7\u6c42\u7684\u65b9\u5f0f\u5b8c\u6210\u58f0\u7eb9\u8bc6\u522b\uff0c\u4f8b\u5982\u63d0\u4f9b\u4e00\u4e2aAPI\u4f9bAPP\u8c03\u7528\uff0c\u7528\u6237\u5728APP\u4e0a\u901a\u8fc7\u58f0\u7eb9\u767b\u5f55\u65f6\uff0c\u628a\u5f55\u97f3\u5230\u7684\u8bed\u97f3\u53d1\u9001\u5230\u540e\u7aef\u5b8c\u6210\u58f0\u7eb9\u8bc6\u522b\uff0c\u518d\u628a\u7ed3\u679c\u8fd4\u56de\u7ed9APP\uff0c\u524d\u63d0\u662f\u7528\u6237\u5df2\u7ecf\u4f7f\u7528\u8bed\u97f3\u6ce8\u518c\uff0c\u5e76\u6210\u529f\u628a\u8bed\u97f3\u6570\u636e\u5b58\u653e\u5728`audio_db`\u6587\u4ef6\u5939\u4e2d\u3002\r\n```shell\r\npython infer_recognition.py\r\n```\r\n\r\n\u8f93\u51fa\u7c7b\u4f3c\u5982\u4e0b\uff1a\r\n```\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:13 - ----------- \u989d\u5916\u914d\u7f6e\u53c2\u6570 -----------\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - audio_db_path: audio_db/\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - configs: configs/ecapa_tdnn.yml\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - model_path: models/EcapaTdnn_Fbank/best_model/\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - record_seconds: 3\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - threshold: 0.6\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:15 - use_gpu: True\r\n[2023-04-02 18:31:20.521040 INFO ] utils:print_arguments:16 - ------------------------------------------------\r\n\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7\r\nW0425 08:30:13.257884 23889 device_context.cc:447] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.6, Runtime API Version: 10.2\r\nW0425 08:30:13.260191 23889 device_context.cc:465] device: 0, cuDNN Version: 7.6.\r\n\u6210\u529f\u52a0\u8f7d\u6a21\u578b\u53c2\u6570\u548c\u4f18\u5316\u65b9\u6cd5\u53c2\u6570\uff1amodels/ecapa_tdnn/model.pth\r\nLoaded \u6c99\u745e\u91d1 audio.\r\nLoaded \u674e\u8fbe\u5eb7 audio.\r\n\u8bf7\u9009\u62e9\u529f\u80fd\uff0c0\u4e3a\u6ce8\u518c\u97f3\u9891\u5230\u58f0\u7eb9\u5e93\uff0c1\u4e3a\u6267\u884c\u58f0\u7eb9\u8bc6\u522b\uff1a0\r\n\u6309\u4e0b\u56de\u8f66\u952e\u5f00\u673a\u5f55\u97f3\uff0c\u5f55\u97f33\u79d2\u4e2d\uff1a\r\n\u5f00\u59cb\u5f55\u97f3......\r\n\u5f55\u97f3\u5df2\u7ed3\u675f!\r\n\u8bf7\u8f93\u5165\u8be5\u97f3\u9891\u7528\u6237\u7684\u540d\u79f0\uff1a\u591c\u96e8\u98d8\u96f6\r\n\u8bf7\u9009\u62e9\u529f\u80fd\uff0c0\u4e3a\u6ce8\u518c\u97f3\u9891\u5230\u58f0\u7eb9\u5e93\uff0c1\u4e3a\u6267\u884c\u58f0\u7eb9\u8bc6\u522b\uff1a1\r\n\u6309\u4e0b\u56de\u8f66\u952e\u5f00\u673a\u5f55\u97f3\uff0c\u5f55\u97f33\u79d2\u4e2d\uff1a\r\n\u5f00\u59cb\u5f55\u97f3......\r\n\u5f55\u97f3\u5df2\u7ed3\u675f!\r\n\u8bc6\u522b\u8bf4\u8bdd\u7684\u4e3a\uff1a\u591c\u96e8\u98d8\u96f6\uff0c\u76f8\u4f3c\u5ea6\u4e3a\uff1a0.920434\r\n```\r\n\r\n\r\n\u540c\u65f6\u8fd8\u63d0\u4f9b\u4e86\u6709GUI\u754c\u9762\u7684\u58f0\u7eb9\u8bc6\u522b\u7a0b\u5e8f\uff0c\u6267\u884c`infer_recognition_gui.py`\u542f\u52a8\uff0c\u70b9\u51fb`\u6ce8\u518c\u97f3\u9891\u5230\u58f0\u7eb9\u5e93`\u6309\u94ae\uff0c\u7406\u89e3\u5f00\u59cb\u8bf4\u8bdd\uff0c\u5f55\u52363\u79d2\u949f\uff0c\u7136\u540e\u8f93\u5165\u6ce8\u518c\u4eba\u7684\u540d\u79f0\uff0c\u4e4b\u540e\u53ef\u4ee5`\u6267\u884c\u58f0\u7eb9\u8bc6\u522b`\u6309\u94ae\uff0c\u7136\u540e\u7acb\u5373\u8bf4\u8bdd\uff0c\u5f55\u52363\u79d2\u949f\u540e\uff0c\u7b49\u5f85\u8bc6\u522b\u7ed3\u679c\u3002`\u5220\u9664\u7528\u6237`\u6309\u94ae\u53ef\u4ee5\u5220\u9664\u7528\u6237\u3002`\u5b9e\u65f6\u8bc6\u522b`\u6309\u94ae\u53ef\u4ee5\u5b9e\u65f6\u8bc6\u522b\uff0c\u53ef\u4ee5\u4e00\u76f4\u5f55\u97f3\uff0c\u4e00\u76f4\u8bc6\u522b\u3002\r\n\r\n<div align=\"center\">\r\n<img src=\"./docs/images/recognition.jpg\" alt=\"\u58f0\u7eb9\u8bc6\u522b\u754c\u9762\">\r\n</div>\r\n\r\n# \u8bf4\u8bdd\u4eba\u65e5\u5fd7\uff08\u5206\u79bb\u8bf4\u8bdd\u4eba\uff09\r\n\r\n\u6267\u884c`infer_speaker_diarization.py`\u7a0b\u5e8f\uff0c\u8f93\u5165\u97f3\u9891\u8def\u5f84\uff0c\u5c31\u53ef\u4ee5\u5206\u79bb\u51fa\u8bf4\u8bdd\u4eba\uff0c\u5e76\u663e\u793a\u7ed3\u679c\uff0c\u5efa\u8bae\u97f3\u9891\u957f\u5ea6\u4e0d\u8981\u4f4e\u4e8e10\u79d2\u3002\u66f4\u591a\u529f\u80fd\u53ef\u4ee5\u67e5\u770b\u8be5\u7a0b\u5e8f\u53c2\u6570\u3002\r\n```shell\r\npython infer_speaker_diarization.py --audio_path=dataset/test_long.wav\r\n```\r\n\r\n\u8f93\u51fa\u7c7b\u4f3c\u5982\u4e0b\uff1a\r\n```\r\n2024-10-10 19:30:40.768 | INFO | mvector.predict:__init__:61 - \u6210\u529f\u52a0\u8f7d\u6a21\u578b\u53c2\u6570\uff1amodels/CAMPPlus_Fbank/best_model/model.pth\r\n2024-10-10 19:30:40.795 | INFO | mvector.predict:__create_index:127 - \u58f0\u7eb9\u7279\u5f81\u7d22\u5f15\u521b\u5efa\u5b8c\u6210\uff0c\u4e00\u5171\u67093\u4e2a\u7528\u6237\uff0c\u5206\u522b\u662f\uff1a['\u6c99\u745e\u91d1', '\u591c\u96e8\u98d8\u96f6', '\u674e\u8fbe\u5eb7']\r\n2024-10-10 19:30:40.796 | INFO | mvector.predict:__load_audio_db:142 - \u6b63\u5728\u52a0\u8f7d\u58f0\u7eb9\u5e93\u6570\u636e...\r\n100%|\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588\u2588| 3/3 [00:00<?, ?it/s]\r\n2024-10-10 19:30:40.798 | INFO | mvector.predict:__create_index:127 - \u58f0\u7eb9\u7279\u5f81\u7d22\u5f15\u521b\u5efa\u5b8c\u6210\uff0c\u4e00\u5171\u67093\u4e2a\u7528\u6237\uff0c\u5206\u522b\u662f\uff1a['\u6c99\u745e\u91d1', '\u591c\u96e8\u98d8\u96f6', '\u674e\u8fbe\u5eb7']\r\n2024-10-10 19:30:40.798 | INFO | mvector.predict:__load_audio_db:172 - \u58f0\u7eb9\u5e93\u6570\u636e\u52a0\u8f7d\u5b8c\u6210\uff01\r\n\u8bc6\u522b\u7ed3\u679c\uff1a\r\n{'speaker': '\u6c99\u745e\u91d1', 'start': 0.0, 'end': 2.0}\r\n{'speaker': '\u964c\u751f\u4eba1', 'start': 4.0, 'end': 7.0}\r\n{'speaker': '\u674e\u8fbe\u5eb7', 'start': 7.0, 'end': 8.0}\r\n{'speaker': '\u6c99\u745e\u91d1', 'start': 9.0, 'end': 12.0}\r\n{'speaker': '\u6c99\u745e\u91d1', 'start': 13.0, 'end': 14.0}\r\n{'speaker': '\u964c\u751f\u4eba1', 'start': 15.0, 'end': 19.0}\r\n```\r\n\r\n\u663e\u793a\u7ed3\u679c\u56fe\u50cf\u5982\u4e0b\uff0c\u53ef\u4ee5\u901a\u8fc7`\u7a7a\u683c`\u952e\u63a7\u5236\u64ad\u653e\u97f3\u9891\uff0c\u70b9\u51fb\u4f4d\u7f6e\u53ef\u4ee5\u8df3\u8f6c\u97f3\u9891\u5230\u6307\u5b9a\u4f4d\u7f6e\uff1a\r\n<div align=\"center\">\r\n<img src=\"./docs/images/speaker_diarization.jpg\" alt=\"\u8bf4\u8bdd\u4eba\u65e5\u5fd7\" width=\"800\">\r\n</div>\r\n\r\n\u9879\u76ee\u540c\u6837\u63d0\u4f9b\u4e86GUI\u754c\u9762\u7684\u7a0b\u5e8f\uff0c\u6267\u884c`infer_speaker_diarization_gui.py`\u7a0b\u5e8f\u3002\u66f4\u591a\u529f\u80fd\u53ef\u4ee5\u67e5\u770b\u8be5\u7a0b\u5e8f\u53c2\u6570\u3002\r\n```shell\r\npython infer_speaker_diarization_gui.py\r\n```\r\n\r\n\u53ef\u4ee5\u6253\u5f00\u8fd9\u6837\u4e00\u4e2a\u9875\u9762\uff0c\u8fdb\u884c\u8bf4\u8bdd\u4eba\u8bc6\u522b\uff1a\r\n\r\n<div align=\"center\">\r\n<img src=\"./docs/images/speaker_diarization_gui.png\" alt=\"\u8bf4\u8bdd\u4eba\u65e5\u5fd7\" width=\"800\">\r\n</div>\r\n\r\n\r\n\u6ce8\u610f\uff1a\u5982\u679c\u8bf4\u8bdd\u4eba\u540d\u5b57\u662f\u4e2d\u6587\u7684\uff0c\u9700\u8981\u8bbe\u7f6e\u5b89\u88c5\u5b57\u4f53\u624d\u80fd\u6b63\u5e38\u663e\u793a\uff0c\u4e00\u822c\u60c5\u51b5\u4e0bWindows\u65e0\u9700\u5b89\u88c5\uff0cUbuntu\u9700\u8981\u5b89\u88c5\u3002\u5982\u679cWindows\u786e\u5b9e\u662f\u7f3a\u5c11\u5b57\u4f53\uff0c\u53ea\u9700\u8981[\u5b57\u4f53\u6587\u4ef6](https://github.com/tracyone/program_font)\u8fd9\u91cc\u4e0b\u8f7d`.ttf`\u683c\u5f0f\u7684\u6587\u4ef6\uff0c\u590d\u5236\u5230`C:\\Windows\\Fonts`\u5373\u53ef\u3002Ubuntu\u7cfb\u7edf\u64cd\u4f5c\u5982\u4e0b\u3002\r\n\r\n1. \u5b89\u88c5\u5b57\u4f53\r\n```shell\r\ngit clone https://github.com/tracyone/program_font && cd program_font && ./install.sh\r\n```\r\n\r\n2. \u6267\u884c\u4e0b\u9762Python\u4ee3\u7801\r\n```python\r\nimport matplotlib\r\nimport shutil\r\nimport os\r\n\r\npath = matplotlib.matplotlib_fname()\r\npath = path.replace('matplotlibrc', 'fonts/ttf/')\r\nprint(path)\r\nshutil.copy('/usr/share/fonts/MyFonts/simhei.ttf', path)\r\nuser_dir = os.path.expanduser('~')\r\nshutil.rmtree(f'{user_dir}/.cache/matplotlib', ignore_errors=True)\r\n```\r\n\r\n\r\n# \u5176\u4ed6\u7248\u672c\r\n - Tensorflow\uff1a[VoiceprintRecognition-Tensorflow](https://github.com/yeyupiaoling/VoiceprintRecognition-Tensorflow)\r\n - PaddlePaddle\uff1a[VoiceprintRecognition-PaddlePaddle](https://github.com/yeyupiaoling/VoiceprintRecognition-PaddlePaddle)\r\n - Keras\uff1a[VoiceprintRecognition-Keras](https://github.com/yeyupiaoling/VoiceprintRecognition-Keras)\r\n\r\n\r\n## \u6253\u8d4f\u4f5c\u8005\r\n\r\n<br/>\r\n<div align=\"center\">\r\n<p>\u6253\u8d4f\u4e00\u5757\u94b1\u652f\u6301\u4e00\u4e0b\u4f5c\u8005</p>\r\n<img src=\"https://yeyupiaoling.cn/reward.png\" alt=\"\u6253\u8d4f\u4f5c\u8005\" width=\"400\">\r\n</div>\r\n\r\n\r\n# \u53c2\u8003\u8d44\u6599\r\n1. https://github.com/PaddlePaddle/PaddleSpeech\r\n2. https://github.com/yeyupiaoling/PaddlePaddle-MobileFaceNets\r\n3. https://github.com/yeyupiaoling/PPASR\r\n4. https://github.com/alibaba-damo-academy/3D-Speaker\r\n5. https://github.com/wenet-e2e/wespeaker\r\n",
"bugtrack_url": null,
"license": "Apache License 2.0",
"summary": "Voice Print Recognition toolkit on Pytorch",
"version": "1.1.1",
"project_urls": {
"Download": "https://github.com/yeyupiaoling/VoiceprintRecognition_Pytorch.git",
"Homepage": "https://github.com/yeyupiaoling/VoiceprintRecognition_Pytorch"
},
"split_keywords": [
"voice",
" pytorch"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "8777fc8719b1fd31309c2f8d34f12d7f39e73a4cf60e78ee373c8313ef6dd292",
"md5": "6ad7f72bd39e5afe363ac5bfd08025de",
"sha256": "2914fac7dba838321ee4c0cd164cf83a0154bc20dfdc9430fb5bdc26abc967b1"
},
"downloads": -1,
"filename": "mvector-1.1.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "6ad7f72bd39e5afe363ac5bfd08025de",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": null,
"size": 73944,
"upload_time": "2024-10-13T12:33:29",
"upload_time_iso_8601": "2024-10-13T12:33:29.121033Z",
"url": "https://files.pythonhosted.org/packages/87/77/fc8719b1fd31309c2f8d34f12d7f39e73a4cf60e78ee373c8313ef6dd292/mvector-1.1.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-10-13 12:33:29",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "yeyupiaoling",
"github_project": "VoiceprintRecognition_Pytorch",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"requirements": [
{
"name": "numpy",
"specs": [
[
">=",
"1.19.2"
]
]
},
{
"name": "tqdm",
"specs": [
[
">=",
"4.59.0"
]
]
},
{
"name": "visualdl",
"specs": [
[
">=",
"2.5.3"
]
]
},
{
"name": "resampy",
"specs": [
[
">=",
"0.2.2"
]
]
},
{
"name": "soundfile",
"specs": [
[
">=",
"0.12.1"
]
]
},
{
"name": "soundcard",
"specs": [
[
">=",
"0.4.2"
]
]
},
{
"name": "pyyaml",
"specs": [
[
">=",
"5.4.1"
]
]
},
{
"name": "scikit-learn",
"specs": [
[
">=",
"1.5.2"
]
]
},
{
"name": "pydub",
"specs": [
[
">=",
"0.25.1"
]
]
},
{
"name": "torchinfo",
"specs": [
[
">=",
"1.7.2"
]
]
},
{
"name": "loguru",
"specs": [
[
">=",
"0.7.2"
]
]
},
{
"name": "yeaudio",
"specs": [
[
">=",
"0.0.6"
]
]
}
],
"lcname": "mvector"
}
JSON
