Starstruk Logo MarkStarstruk

v1.3.15

modelscope/FunASRv1.3.15Jul 17, 2026by github-actions[bot]

AI Summary

FunASR v1.3.15 focuses on production fixes to improve streaming reliability, fix Fun-ASR-Nano compatibility, and refine CLI and text processing behaviors.

Key Highlights

  • Enhanced streaming stability by bounding VAD frame buffers and fixing duplicate dynamic-silence state updates.
  • Resolved Fun-ASR-Nano issues including stereo audio preprocessing and NPU autocast device selection.
  • Improved CLI functionality with correct hotword routing and explicit model-hub selection.
  • Replaced `editdistance` with RapidFuzz for maintained error-rate metrics and Unicode handling.

New Features

  • Added explicit model-hub selection to the CLI.
  • Added post-processing hotword correction to `AutoModel.generate()`.
  • Exposed keepalive controls for realtime WebSocket sessions.
  • Accept float32 runtime dtype.

Full Release Notes

FunASR v1.3.15 rolls up the production fixes merged since v1.3.14, with a focus on long-running streaming sessions, Fun-ASR-Nano compatibility, and predictable CLI/text behavior.

## Streaming and deployment reliability

- Bound streaming VAD frame buffers so long sessions do not retain audio indefinitely. ([#3236](https://github.com/modelscope/FunASR/pull/3236))
- Remove duplicate dynamic-silence state updates that could produce inconsistent segment boundaries. ([#3240](https://github.com/modelscope/FunASR/pull/3240))
- Bound realtime WebSocket session state, accept float32 runtime dtype, and expose keepalive controls. ([#3214](https://github.com/modelscope/FunASR/pull/3214), [#3204](https://github.com/modelscope/FunASR/pull/3204), [#3121](https://github.com/modelscope/FunASR/pull/3121))

## Fun-ASR-Nano fixes

- Correct stereo-audio preprocessing and NPU autocast device selection. ([#3102](https://github.com/modelscope/FunASR/pull/3102), [#3113](https://github.com/modelscope/FunASR/pull/3113))
- Make CTC batch fallback robust and disable timestamps cleanly when a checkpoint has no CTC weights. ([#3187](https://github.com/modelscope/FunASR/pull/3187), [#3211](https://github.com/modelscope/FunASR/pull/3211))
- Warn before a single vLLM request exceeds the supported audio window. ([#3033](https://github.com/modelscope/FunASR/pull/3033))

## CLI and text processing

- Route Paraformer CLI hotwords correctly and add post-processing hotword correction to `AutoModel.generate()`. ([#3237](https://github.com/modelscope/FunASR/pull/3237), [#2959](https://github.com/modelscope/FunASR/pull/2959))
- Add explicit model-hub selection to the CLI. ([#3046](https://github.com/modelscope/FunASR/pull/3046))
- Prevent SenseVoice special-token strings from being mistaken for batched text and handle Unicode punctuation consistently. ([#3110](https://github.com/modelscope/FunASR/pull/3110), [#3238](https://github.com/modelscope/FunASR/pull/3238))
- Replace `editdistance` with RapidFuzz for maintained, guarded error-rate metrics.

## Upgrade

```bash
pip install --upgrade funasr==1.3.15
```

**Full changelog:** https://github.com/modelscope/FunASR/compare/v1.3.14...v1.3.15