PH · PRODUCT_HUNT// 4h agoOPENSOURCE RELEASE

Xiaomi releases MiMo-V2.5 open-source voice model suite

MiMo-V2.5 is an 8B parameter open-source speech model suite from Xiaomi that provides high-accuracy ASR and TTS capabilities. It excels at transcribing Mandarin, English, and eight Chinese dialects, featuring native support for mid-sentence code-switching and complex song lyrics transcription.

// ANALYSIS

Xiaomi is moving beyond generic transcription to solve difficult edge cases like multi-dialect support and music.

–Native prosody-based punctuation eliminates the need for separate post-processing models.
–Superior performance over Whisper large-v3 in English (5.73% vs 7.44% WER).
–Optimized for "in-the-wild" audio including heavy background noise and musical accompaniment.
–8B parameter size balances accuracy with the ability to run on consumer-grade hardware.

// TAGS

mimo-v2-5-voicespeechasrttsopen-sourcexiaomi

DISCOVERED

4h ago

2026-04-25

PUBLISHED

9h ago

2026-04-25

RELEVANCE

8/ 10