feat: add FunASR audio transcription integration

[Akash504-ai]

Related Issues

• Closes Feature: Add FunASR as audio transcription component (13x faster, speaker diarization) #3373

Proposed Changes

Adds a new funasr-haystack integration with a FunASRTranscriber component.

Features:

• Audio transcription using the FunASR Python SDK (AutoModel)
• CPU and GPU support
• Timestamp metadata support
• Optional speaker diarization support
• Optional SenseVoice-style emotion/event tag extraction
• Lazy model loading through warm_up()
• Serialization and deserialization support
• Haystack Document output compatible with pipelines

The implementation follows the existing Haystack integration structure and introduces a standalone integration under integrations/funasr.

How did you test it?

• Added unit tests using a mocked FunASR SDK
• Added an opt-in integration test for real audio transcription
• Verified serialization/deserialization behavior
• Verified pipeline round-trip serialization
• Ran Ruff checks and formatting
• Ran type checking with mypy
• Ran documentation generation

Notes for the reviewer

The implementation currently uses the official FunASR Python SDK (AutoModel) rather than the OpenAI-compatible endpoint.

I would especially appreciate feedback on:

• output schema for speaker diarization metadata
• emotion/event tag handling
• whether the SDK approach is preferred over the OpenAI-compatible API approach

Checklist

• I have read the contributors guidelines and code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used a conventional commit type for the PR title

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	funasr@​1.3.9

View full report