# XiaomiMiMo/MiMo-Audio

MiMo-Audio: Audio Language Models are Few-Shot Learners

Repository: https://github.com/XiaomiMiMo/MiMo-Audio
Canonical: https://ross.abutalabs.com/products/mimo-audio
Homepage: https://xiaomimimo.github.io/MiMo-Audio-Demo/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-06-17T09:51:01+00:00

## Health v2 (maintenance only)
Score: 57/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 88, release rhythm 35, longevity 24
- inputs: {"age_days": 349, "days_push": 77, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1077, forks 107 (observed 2026-08-28T04:03:29.557089+00:00)

## What it is
Xiaomi's open-source 7B audio language model family (Base and Instruct) plus a 1.2B RVQ audio tokenizer, pretrained on 100M+ hours of audio to enable few-shot generalization across audio tasks. It jointly models text and audio tokens for understanding, speech generation, TTS, voice conversion, and speech editing.

## Use cases
- build an audio understanding model that answers questions about speech and sound
- text-to-speech and instruct-TTS with natural voices
- speech continuation for generating talk shows or recitations
- voice conversion and style transfer with few examples
- speech-to-text transcription and audio reasoning benchmarks
- few-shot audio task generalization without task-specific fine-tuning

## When to choose
- you need a single open-source model covering both audio understanding and generation
- you want SOTA open-source performance on audio benchmarks like MMAU or MMSU
- you need few-shot generalization to audio tasks without fine-tuning
- you want an Apache-2.0 licensed audio LLM with released checkpoints and eval suite

## When to avoid
- you need lightweight on-device audio processing rather than a 7B GPU model
- you only need simple ASR where Whisper-class models suffice
- you lack GPU resources for large model inference
- you need production hardening rather than a research codebase

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, audio-processing, speech-recognition, tts, llm-inference
- domain: artificial-intelligence, speech-processing, large-language-models
- platform: python
- tags: audio-language-model, few-shot-learning, audio-tokenizer, rvq, speech-generation, instruction-tuning, 7b-model, multimodal, audio, natural-language-processing, gpu, linux

## Member repositories
- XiaomiMiMo/MiMo-Audio (main) score 57

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:29.557089+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:53:13.611595+00:00, confidence not recorded.
  - readme: https://github.com/XiaomiMiMo/MiMo-Audio (fetched 2026-08-28T04:03:29.557089+00:00, sha 7346235026dc)
  - homepage: https://xiaomimimo.github.io/MiMo-Audio-Demo/ (fetched 2026-08-29T12:54:41.751501+00:00, sha 6c849607bcb2)
- Data as of 2026-08-30T08:39:29.467469+00:00.