FunASR v1.4.16: Mixed-device Submodels and Native Transformers Guides

September 18, 2026 · Stable FunASR release

v1.4.16 keeps explicitly configured VAD, punctuation, and speaker submodels on their own devices across repeated inference and runtime overrides, while adding clearer punctuation and native Transformers guidance.

Install

python -m pip install -U "funasr==1.4.16"

The public distributions are on PyPI 1.4.16. The wheel SHA-256 is f95943f6a8111349a1b08907363b048a49e9cb9caa1e3afd8c92970d08bf450d; the sdist SHA-256 is 1e2f8e3887fbc4b164cce40f73bf1324907da6ab59d1e8bf2add827f534c69e7.

What changed

• Explicit device placement in vad_kwargs, punc_kwargs, and spk_kwargs stays stable across repeated calls.
• Runtime device overrides no longer mutate loaded models or caller-owned submodel configuration.
• Speaker inference uses its own model configuration; English and Chinese guides explain when CT-Transformer punctuation postprocessing is needed and how to validate native Transformers paths.

Signed source and verifiable assets

The signed v1.4.16 tag resolves exactly to 904cd18681b8083de5e1039bd0ecebc4f49ede60. The Release has exactly 12 assets: two Python distributions, nine independently versioned runtimes from runtime-llamacpp-v0.2.6, and one manifest.

• Python: funasr-1.4.16-py3-none-any.whl, funasr-1.4.16.tar.gz
• Linux: funasr-llamacpp-linux-arm64.tar.gz, funasr-llamacpp-linux-x64.tar.gz, funasr-llamacpp-linux-x64-avx2.tar.gz, funasr-llamacpp-linux-x64-vulkan.tar.gz
• macOS: funasr-llamacpp-macos-arm64.tar.gz
• Windows: funasr-llamacpp-windows-x64.zip, funasr-llamacpp-windows-x64-avx2.zip, funasr-llamacpp-windows-x64-cuda.zip, funasr-llamacpp-windows-x64-vulkan.zip
• Verification: SHA256SUMS

The nine native runtimes do not contain this Python fix. Verify the manifest first, then validate on target hardware with representative audio.