# L3AC
This repository contains the implementation of L3AC, a lightweight audio codec based on a single quantizer,
introduced in the paper titled "L3AC: Towards a Lightweight and Lossless Audio Codec".
[Paper](https://arxiv.org/abs/2504.04949)
[Model Weights](https://huggingface.co/zhai-lw/L3AC)
<figure class="image">
<img src="./bubble_chart.svg" alt="Comparison of various audio codec">
<figcaption>Comparison of various audio codec</figcaption>
</figure>
## install
```
pip install l3ac
```
### demo
Firstly, make sure you have installed the librosa package to load the example audio file. You can install it using pip:
```
pip install librosa
```
Then, you can use the following code to load a sample audio file, encode it using the L3AC model, and decode it back
to audio. The code also calculates the mean squared error (MSE) between the original and generated audio.
```python
import librosa
import torch
import l3ac
all_models = l3ac.list_models()
print(f"Available models: {all_models}")
MODEL_USED = '1kbps'
codec = l3ac.get_model(MODEL_USED)
print(f"loaded codec({MODEL_USED}) and codec sample rate: {codec.config.sample_rate}")
sample_audio, sample_rate = librosa.load(librosa.example("libri1"))
sample_audio = sample_audio[None, :]
print(f"loaded sample audio and audio sample_rate :{sample_rate}")
sample_audio = librosa.resample(sample_audio, orig_sr=sample_rate, target_sr=codec.config.sample_rate)
codec.network.cuda()
codec.network.eval()
with torch.inference_mode():
audio_in = torch.tensor(sample_audio, dtype=torch.float32, device='cuda')
_, audio_length = audio_in.shape
print(f"{audio_in.shape=}")
q_feature, indices = codec.encode_audio(audio_in)
audio_out = codec.decode_audio(q_feature) # or
# audio_out = codec.decode_audio(indices=indices['indices'])
generated_audio = audio_out[:, :audio_length].detach().cpu().numpy()
mse = ((sample_audio - generated_audio) ** 2).mean().item()
print(f"codec({MODEL_USED}) mse: {mse}")
```
### available models
| config_name | Sample rate(Hz) | tokens/s | Codebook size | Bitrate(bps) |
|-------------|-----------------|----------|---------------|--------------|
| 0k75bps | 16,000 | 44.44 | 117,649 | 748.6 |
| 1kbps | 16,000 | 59.26 | 117,649 | 998.2 |
| 1k5bps | 16,000 | 88.89 | 117,649 | 1497.3 |
| 3kbps | 16,000 | 166.67 | 250,047 | 2988.6 |
Raw data
{
"_id": null,
"home_page": null,
"name": "l3ac",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.11",
"maintainer_email": null,
"keywords": "artificial intelligence, audio codec, deep learning, pytorch",
"author": null,
"author_email": "Linwei Zhai <zhai_lw@outlook.com>",
"download_url": "https://files.pythonhosted.org/packages/84/ee/dba24a2924ce02f0233cf3ba6757e0c649251ce2aad92e4cae2220a2e189/l3ac-1.0.tar.gz",
"platform": null,
"description": "# L3AC\n\nThis repository contains the implementation of L3AC, a lightweight audio codec based on a single quantizer,\nintroduced in the paper titled \"L3AC: Towards a Lightweight and Lossless Audio Codec\".\n\n[Paper](https://arxiv.org/abs/2504.04949)\n\n[Model Weights](https://huggingface.co/zhai-lw/L3AC)\n\n\n<figure class=\"image\">\n <img src=\"./bubble_chart.svg\" alt=\"Comparison of various audio codec\">\n <figcaption>Comparison of various audio codec</figcaption>\n</figure>\n\n## install\n\n```\npip install l3ac\n```\n\n### demo\n\nFirstly, make sure you have installed the librosa package to load the example audio file. You can install it using pip:\n\n```\npip install librosa\n```\n\nThen, you can use the following code to load a sample audio file, encode it using the L3AC model, and decode it back\nto audio. The code also calculates the mean squared error (MSE) between the original and generated audio.\n\n```python\nimport librosa\nimport torch\nimport l3ac\n\nall_models = l3ac.list_models()\nprint(f\"Available models: {all_models}\")\n\nMODEL_USED = '1kbps'\ncodec = l3ac.get_model(MODEL_USED)\nprint(f\"loaded codec({MODEL_USED}) and codec sample rate: {codec.config.sample_rate}\")\n\nsample_audio, sample_rate = librosa.load(librosa.example(\"libri1\"))\nsample_audio = sample_audio[None, :]\nprint(f\"loaded sample audio and audio sample_rate :{sample_rate}\")\n\nsample_audio = librosa.resample(sample_audio, orig_sr=sample_rate, target_sr=codec.config.sample_rate)\n\ncodec.network.cuda()\ncodec.network.eval()\nwith torch.inference_mode():\n audio_in = torch.tensor(sample_audio, dtype=torch.float32, device='cuda')\n _, audio_length = audio_in.shape\n print(f\"{audio_in.shape=}\")\n q_feature, indices = codec.encode_audio(audio_in)\n audio_out = codec.decode_audio(q_feature) # or\n # audio_out = codec.decode_audio(indices=indices['indices'])\n generated_audio = audio_out[:, :audio_length].detach().cpu().numpy()\n\nmse = ((sample_audio - generated_audio) ** 2).mean().item()\nprint(f\"codec({MODEL_USED}) mse: {mse}\")\n```\n\n### available models\n\n| config_name | Sample rate(Hz) | tokens/s | Codebook size | Bitrate(bps) |\n|-------------|-----------------|----------|---------------|--------------|\n| 0k75bps | 16,000 | 44.44 | 117,649 | 748.6 |\n| 1kbps | 16,000 | 59.26 | 117,649 | 998.2 |\n| 1k5bps | 16,000 | 88.89 | 117,649 | 1497.3 |\n| 3kbps | 16,000 | 166.67 | 250,047 | 2988.6 |\n",
"bugtrack_url": null,
"license": null,
"summary": "L3AC: A nerual audio codec with one quantizer",
"version": "1.0",
"project_urls": {
"DOWNLOAD": "https://huggingface.co/zhai-lw/L3AC",
"Documentation": "https://github.com/zhai-lw/L3AC#readme",
"Issues": "https://github.com/zhai-lw/L3AC/issues",
"Repository": "https://github.com/zhai-lw/L3AC.git"
},
"split_keywords": [
"artificial intelligence",
" audio codec",
" deep learning",
" pytorch"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "94a46a6b42f327a2d406f5411c51769617d1c5f156eb94f4a341204687ced31e",
"md5": "9237381ed6b5fd65a21d67d8f02d8e5e",
"sha256": "77be7b25d164cb00563ab5c8d33184b82466c9354323ac93000277ae225102ba"
},
"downloads": -1,
"filename": "l3ac-1.0-py3-none-any.whl",
"has_sig": false,
"md5_digest": "9237381ed6b5fd65a21d67d8f02d8e5e",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.11",
"size": 21952,
"upload_time": "2025-08-15T12:24:50",
"upload_time_iso_8601": "2025-08-15T12:24:50.415292Z",
"url": "https://files.pythonhosted.org/packages/94/a4/6a6b42f327a2d406f5411c51769617d1c5f156eb94f4a341204687ced31e/l3ac-1.0-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "84eedba24a2924ce02f0233cf3ba6757e0c649251ce2aad92e4cae2220a2e189",
"md5": "a39cf6438f2c428b0c584045202eb980",
"sha256": "d596faf1e2dd177b3218cb24e5623de60bbf665047a3c603aa21cf84400cbde5"
},
"downloads": -1,
"filename": "l3ac-1.0.tar.gz",
"has_sig": false,
"md5_digest": "a39cf6438f2c428b0c584045202eb980",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.11",
"size": 25724,
"upload_time": "2025-08-15T12:24:49",
"upload_time_iso_8601": "2025-08-15T12:24:49.011632Z",
"url": "https://files.pythonhosted.org/packages/84/ee/dba24a2924ce02f0233cf3ba6757e0c649251ce2aad92e4cae2220a2e189/l3ac-1.0.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-08-15 12:24:49",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "zhai-lw",
"github_project": "L3AC#readme",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "l3ac"
}
JSON
