Lab-et/amharic-corpus 数据集
合成阿姆哈拉语通话语音语料，面向 ASR 与 TTS。
为什么重要
为低资源阿姆哈拉语 ASR 预训练和领域适配提供可直接获取的合成语料。
阅读原文huggingface.co
AI 解读（根据原文生成）
Lab-et/amharic-corpus 是 Hugging Face 上的阿姆哈拉语合成语音数据集，包含 146,359 个片段、约 408.5 小时音频，16 kHz、单声道、16-bit WAV。语料由 lab.et 的 OmniVoice Amharic TTS 生成，模拟银行、交通、电信和服务场景的埃塞俄比亚呼叫中心对话，含 62 个合成说话人身份。数据以 Parquet shards 提供，字段包括 audio、text、duration、speaker_id 等，许可证为 cc-by-4.0，需申请访问为否。
要点（AI 据原文整理，数字均已与原文核对）
- 146,359 clips · 408.5 hours · Amharic (አማርኛ) · 16 kHz · synthetic, studio-quality channel
- The audio simulates Ethiopian call-center conversations across banking, transport, telecom and everyday service scenarios, rendered with 62 distinct synthetic speaker identities (mixed genders).
- Sample rate 16 kHz, mono, 16-bit WAV
- This release is the clean channel of the corpus: near-field, studio-quality speech.
- A telephony-degraded channel variant exists but is not part of this release.
- No real customer data was used at any stage: the dialogue text is synthetic, written by LLMs for training scenarios, then spoken by OmniVoice.
