ggfox00000/stt-mls-test

MLS法语测试分割数据集是Multilingual LibriSpeech (MLS)的一部分，专门用于法语自动语音识别(ASR)的基准测试。数据集包含2426条法语语音片段，音频格式为16 kHz单声道FLAC，总时长为10.07小时。每条数据包含音频文件、转录文本、说话者ID、书籍ID、源URL以及音频片段的时间信息。数据集主要用于评估法语ASR模型的性能（如WER/CER），并支持直接通过Hugging Face的`load_dataset`加载。数据来源于LibriVox的公共领域有声读物，采用CC-BY-4.0许可证。

The MLS French test split is a part of the Multilingual LibriSpeech (MLS) dataset, specifically designed for benchmarking French automatic speech recognition (ASR). It contains 2426 French speech utterances in 16 kHz mono FLAC format, with a total duration of 10.07 hours. Each entry includes audio files, transcriptions, speaker IDs, book IDs, source URLs, and timing information for audio segments. The dataset is primarily used to evaluate the performance of French ASR models (e.g., WER/CER) and supports direct loading via Hugging Faces `load_dataset`. The data originates from LibriVox public domain audiobooks and is licensed under CC-BY-4.0.