🎵AudioCodecBench Documentation

飞书用户9639

10月15日修改

The usage documentation consists of five parts: Environment Build, User Guides, Advanced Guides, Dataset Download, and Dataset Script Usage.​

1、Environment Build: quickly build a virtual environment, pull projects, and download dependencies.​

2、User Guides: Includes quick tutorials for the reconstruction task, ID sensitivity, perplexity (PPL), and probe task. The reconstruction task evaluates the audio reconstruction fidelity of the codec, the ID sensitivity task evaluates the stability of the codebook ID, the perplexity (PPL) task explores the impact of different token sequences on LM modeling, and the probe task evaluates the performance of representations in downstream tasks.​

3、Advanced Guides: This section describes how to add your own codecs for task testing. It includes adding and initializing codecs, configuring the PPL config script, configuring downstream task configs, and using related datasets.​

4、Dataset Download: The benchmark-related dataset has been processed and open-sourced on huggingface. Download address: https://huggingface.co/datasets/LeBeGut/AudioCodecBench

5、Dataset Script Usage: Since some datasets are large and warehouse storage space is limited, we provide a complete dataset processing script. See the specific path in this chapter. Instructions for use are given in the code comments.​

Environment Build

User Guides

Advanced Guides

Dataset Download

🎵AudioCodecBench Documentation​

🎵AudioCodecBench Documentation