WenetSpeech-Wu-Bench

# WenetSpeech-Wu Bench

We introduce WenetSpeech-Wu-Bench, the first publicly available, manually curated benchmark for Wu dialect speech processing, covering ASR, Wu-to-Mandarin AST, speaker attributes, emotion recognition, TTS, and instruct TTS, and providing a unified platform for fair evaluation.

- **ASR:** Wu dialect ASR (9.75 hour, including Shanghainese, Suzhounese, and Mandarin code-mixed speech). Evaluated by CER.
- **Wu→Mandarin AST:** Speech translation from Wu dialects to Mandarin (3k utterances, 4.4h). Evaluated by BLEU.
- **Speaker Attributes & Emotion:** Speaker gender/age prediction and emotion recognition on Wu dialect. Evaluated by classification accuracy.
- **TTS:** Wu dialect TTS with speaker prompting (242 sentences, 12 speakers). Evaluated by speaker similarity, CER, and MOS.
- **Instruct TTS:** Instruction-following TTS with prosodic and emotional control. Evaluated by automatic accuracy and subjective MOS.

# WenetSpeech-Wu Bench
我们推出了WenetSpeech-Wu Bench，这是首个公开可用、经人工精选整理的吴语语音处理基准评测集，涵盖自动语音识别（ASR, Automatic Speech Recognition）、吴语至普通话语音翻译（AST, Automatic Speech Translation）、说话人属性识别、情感识别、文本转语音（TTS, Text-to-Speech）以及指令式文本转语音任务，并为公平评测提供了统一的平台。

- **自动语音识别（ASR, Automatic Speech Recognition）**：面向吴语的自动语音识别任务（数据集总时长9.75小时，涵盖上海话、苏州话及普通话与吴语的语码混合语音），以字符错误率（CER, Character Error Rate）作为评测指标。
- **吴语→普通话语音翻译（AST, Automatic Speech Translation）**：将吴语语音翻译为普通话的任务（包含3000条语音片段，总时长4.4小时），以双语评估替换分数（BLEU, Bilingual Evaluation Understudy）作为评测指标。
- **说话人属性与情感识别**：针对吴语语音的说话人性别/年龄预测与情感识别任务，以分类准确率作为评测指标。
- **文本转语音（TTS, Text-to-Speech）**：支持说话人提示的吴语文本转语音任务（包含242条句子，覆盖12位说话人），以说话人相似度、字符错误率以及平均意见得分（MOS, Mean Opinion Score）作为评测指标。
- **指令式文本转语音（Instruct TTS）**：支持韵律与情感控制的指令跟随型文本转语音任务，以自动准确率与主观平均意见得分作为评测指标。