Starstruk Logo MarkStarstruk

v1.3.11

kubernetes-sigs/agent-sandboxv1.3.11Jun 20, 2026by github-actions[bot]

AI Summary

This release adds llama.cpp/GGUF runtime support for CPU/edge devices, introduces a built-in FSMN-VAD for single-binary speech segmentation, and enhances audio input flexibility via miniaudio.

Key Highlights

  • Add llama.cpp / GGUF runtime (Fun-ASR-Nano, SenseVoice, Paraformer)
  • Built-in FSMN-VAD (--vad) for speech segmentation without Python
  • Accept any audio input (wav/mp3/flac) via miniaudio
  • B1 packaging for one-command download and standalone conversion

New Features

  • llama.cpp / GGUF runtime support
  • Built-in FSMN-VAD
  • miniaudio audio input support
  • B1 packaging (one-command download)
  • In-binary transcription text printing

Full Release Notes

## What's Changed
* docs: python wss server now supports multiple concurrent clients by @LauraGPT in https://github.com/modelscope/FunASR/pull/2985
* docs: make README quickstart runnable and output truthful by @LauraGPT in https://github.com/modelscope/FunASR/pull/2986
* docs: fix README streaming example (runnable + actually streams) by @LauraGPT in https://github.com/modelscope/FunASR/pull/2987
* Add llama.cpp / GGUF runtime (Fun-ASR-Nano, SenseVoice, Paraformer) by @LauraGPT in https://github.com/modelscope/FunASR/pull/2988
* docs: link llama.cpp / GGUF (CPU/edge) runtime from Deploy section by @LauraGPT in https://github.com/modelscope/FunASR/pull/2991
* ci: auto-create GitHub Release on version tag push by @LauraGPT in https://github.com/modelscope/FunASR/pull/2995
* docs: CPU benchmark vs whisper.cpp (Chinese ASR) by @LauraGPT in https://github.com/modelscope/FunASR/pull/2992
* feat: accept any audio input (any rate/channels, wav/mp3/flac) via miniaudio by @LauraGPT in https://github.com/modelscope/FunASR/pull/2994
* feat: built-in FSMN-VAD (--vad) — single-binary speech segmentation, no Python at runtime by @LauraGPT in https://github.com/modelscope/FunASR/pull/2998
* fix: FSMN-VAD review findings (MSVC M_PI, short-audio guard, tensor validation) by @LauraGPT in https://github.com/modelscope/FunASR/pull/2999
* feat: B1 packaging — one-command download, standalone convert, CI-friendly CMake by @LauraGPT in https://github.com/modelscope/FunASR/pull/3000
* docs: build note for funasr-common (A1 follow-up) by @LauraGPT in https://github.com/modelscope/FunASR/pull/2996
* ci: cross-platform prebuilt binaries for the llama.cpp runtime by @LauraGPT in https://github.com/modelscope/FunASR/pull/3001
* fix: B1 script portability (HF CLI fallback + friendly missing-dep error) by @LauraGPT in https://github.com/modelscope/FunASR/pull/3002
* test: numerical regression harness (frozen golden vs ggml/VAD/CIF/CTC output) by @LauraGPT in https://github.com/modelscope/FunASR/pull/3003
* feat: print transcription text in the binaries (in-binary detok) by @LauraGPT in https://github.com/modelscope/FunASR/pull/3004
* fix: detok review findings (null vocab guard + utf-8 tokens read) by @LauraGPT in https://github.com/modelscope/FunASR/pull/3005
* fix(glm_asr): warn when vLLM dtype=fp16 (degraded output) by @SuperMarioYL in https://github.com/modelscope/FunASR/pull/2993
* fix(glm_asr): honor sampling params in vLLM generate() by @SuperMarioYL in https://github.com/modelscope/FunASR/pull/2997


**Full Changelog**: https://github.com/modelscope/FunASR/compare/v1.3.10...v1.3.11