Skip to content

Feature Request: Add SenseVoice/FunASR for faster on-device ASR #7571

@LauraGPT

Description

@LauraGPT

Feature Request

Omi is an impressive AI wearable that captures and processes audio. Suggesting SenseVoice as an ASR option — its speed and compact size make it ideal for wearable AI.

Why SenseVoice for Omi?

  • Non-autoregressive — instant results, no waiting for sequential decoding
  • 5x faster than Whisper — critical for always-on wearable processing
  • 234M params — much lighter than Whisper large-v3 (1.5B), suitable for edge/mobile
  • 50+ languages — single model for any language the user speaks
  • Emotion detection — detects speaker emotions, valuable for context-aware AI
  • Audio events — identifies laughter, music, applause (contextual awareness)
  • ONNX available — deployable via Sherpa-ONNX on mobile (iOS/Android)

Mobile/Edge deployment options

  1. Sherpa-ONNX — native iOS/Android SDKs available, runs SenseVoice on-device
  2. Server-side — OpenAI-compatible API via funasr-server for cloud processing

Battery/Performance benefits

Non-autoregressive inference means:

Metadata

Metadata

Assignees

No one assigned

    Labels

    p3Priority: Backlog (score <14)

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions