MiMo V2.5 ASR
V2.5 ASR targets transcription across Mandarin, English, Chinese dialects, mixed-language speech and noisy recordings. Published weights are available alongside the hosted service.
Punctuation is generated with the transcript
The card describes native punctuation, lyrics recognition and support for overlapping speakers. It recommends automatic language selection for code-switched speech, with optional Chinese or English tags.MiMo-V2.5-ASR official model card
Check difficult speech rather than clean demos
Practical guidance. Use recordings with the accents, background sound and specialist names expected in the application. Review omissions as well as substitutions. For meetings, do not assume that transcription of overlapping voices automatically provides reliable speaker identities.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | mimo-v2.5-asrMiMo-V2.5-ASR |
| Context window | 8,192 tokens Hosted ASR catalogue lists an 8K context. MiMo current model catalogue |
| Maximum output | 2,048 tokens Hosted ASR catalogue lists a 2K output limit. MiMo current model catalogue |
| Licence | MIT.MiMo-V2.5-ASR official model card |
| Dialect focus | The card names Wu, Cantonese, Hokkien and Sichuanese among its target dialects.MiMo-V2.5-ASR official model card |
| Local dependency | Local inference requires the MiMo-Audio-Tokenizer as well as the ASR weights.MiMo-V2.5-ASR official model card |
Related models
- MiMo Audio 7B Instruct
Xiaomi · MiMo Audio
An instruction-tuned audio model for spoken dialogue, audio understanding and text-to-speech.
Launch date not confirmedDownloadable - MiMo V2.5 TTS
Xiaomi · MiMo V2.5 Speech
The standard V2.5 TTS model turns text into speech using Xiaomi's built-in voices.
Launch date not confirmedAPI