v1.4.26

Final Publication Acceptance

GitHub Release and PyPI 1.4.26 are published. Both official PyPI distributions were downloaded and match GitHub by SHA-256 and size:

• funasr-1.4.26-py3-none-any.whl: 1015738 bytes; SHA-256 104e3704fed0ecfd6c08836468de147a2c6f05745281093b8a6dbe341d1ce831.
• funasr-1.4.26.tar.gz: 1086172 bytes; SHA-256 763c7d860a49afd17e4b9c0636ff43f03f6519e68bd0a09a25d68318091e2cf8.

FunASR 1.4.26

Source: cdd82b5cbfd05015f8fb2b167d61f25142c3a9e8 / tag v1.4.26.

Fixes

• #3874: response_format=text now returns the bare HTTP transcript instead of a JSON-quoted string.
• #3872 / #3883: retain word separators across English and Korean VAD segments while preserving CJK and event-only behavior. The original author's commit is retained.

The independent ONNX/Torch abbreviation and batch-padding timestamp fixes are not shipped by this root wheel. Their package releases are separate. Open #3861, #3862, #3867 and the new hotword PRs are not claimed resolved.

Verification Scope

• #3874 previously passed 32 actual HTTP/SDK cases and 213 related CPU cases; its source bytes remain unchanged here.
• #3883 passed 246 unique selected CPU cases across the initial 230 passes and completion of six optional-dependency cases plus ten actual installed standalone-wheel cases. All original skipped and negative runs remain preserved. NumPy hosted CI has 421 passes and 27 CUDA-only skips per version, not GPU acceptance.
• Release/docs candidate checks: {'contract-green': {'testcase': 53, 'failure': 0, 'error': 0, 'skipped': 0}, 'static-remainder': {'testcase': 104, 'failure': 0, 'error': 0, 'skipped': 2}, 'installed-behavior-candidate': {'testcase': 111, 'failure': 0, 'error': 0, 'skipped': 0}, 'site-boundary-integration': {'testcase': 69, 'failure': 0, 'error': 0, 'skipped': 0}, 'site-manifest-integration': {'testcase': 21, 'failure': 0, 'error': 0, 'skipped': 0}}. Actual installed root wheel final counts: {'testcase': 111, 'failure': 0, 'error': 0, 'skipped': 0}. Standalone runtime paths are explicitly deselected from root-wheel tests. All 439 package files and 13 entrypoints match accepted source.
• Root model/GPU-runtime bytes are unchanged from 1.4.25 and remain bound to its accepted prior L20Z evidence. No new GPU execution, pretrained ASR-quality, full-suite or Nano result is claimed.
• Head, merged-main and final-tag hosted CI and actual checkout trees are separately accepted. Final Python wheel and sdist are newly downloaded and checked for metadata, exact source bytes, archive integrity, strict Twine, private offline installation and CLI behavior.
• Ten unchanged runtime archives reuse complete cache bytes, freshly rehashed against the new assets and original runtime release with archive-integrity checks. They were not rebuilt or re-executed. SHA256SUMS covers all 12 payloads.
• The 1.4.24 warning still applies: checkpoints trained with both norm_means=True and norm_vars=True need recognition-quality reevaluation. norm_vars=False remains the default.
• No production upgrade or adoption/growth claim.

What's Changed

• fix(runtime): release ONNX non-WAV audio loading as 0.4.10 by @LauraGPT in #3876
• fix(server): return plain text for response_format=text by @David-Wu1119 in #3874
• fix(runtime): merge spelled-out acronyms into one word in funasr_onnx and funasr_torch sentence_postprocess by @tanishqbaweja in #3869
• fix(runtime): trim batch padding from CIF peaks before funasr_onnx and funasr_torch Paraformer timestamps by @tanishqbaweja in #3871
• fix(postprocess): preserve VAD word boundaries with current-main CI by @LauraGPT in #3883
• fix(postprocess): keep a space between Latin VAD segments in rich transcription by @David-Wu1119 in #3872
• Release FunASR 1.4.26 by @LauraGPT in #3884

New Contributors

• @David-Wu1119 made their first contribution in #3874

Full Changelog: v1.4.25...v1.4.26

Runtime downloads

This Python release pairs with the current prebuilt llama.cpp / GGUF runtime release: runtime-llamacpp-v0.2.6.

The same verified runtime assets are attached directly to this Python release so users can find the package and self-contained llama-funasr-* binaries in one place.

Platform	Asset	SHA-256
Linux arm64	funasr-llamacpp-linux-arm64.tar.gz	7bca29cfa3c9a08e235a62212ca9e00f6656e59a8f07078966a2bfda1e5aa1f9
Linux x64 AVX2	funasr-llamacpp-linux-x64-avx2.tar.gz	aaebc5470f846ce915200b35d6e9f9bd0a0d3ed399d39e49bdeb7a1f1782bc70
Linux x64 Vulkan	funasr-llamacpp-linux-x64-vulkan.tar.gz	f02d41e98e9d4041f0896661007193810f025484d2175958f7c1313d5c90ec46
Linux x64 portable	funasr-llamacpp-linux-x64.tar.gz	779967de1c528c2be966bcc47f246e7d3e6fcdb748d9491263062f4120f35e52
macOS arm64	funasr-llamacpp-macos-arm64.tar.gz	bda59474202b887190f59d25b7b42c714469efae71276072c12fa0a38de68792
Windows x64 AVX2	funasr-llamacpp-windows-x64-avx2.zip	062cda8fefadd31c3e811227116daccf448a8520f4b0bb168d225c896e65ebbd
Windows x64 CUDA Blackwell (sm_120)	funasr-llamacpp-windows-x64-cuda-blackwell.zip	e32961a753f40888182f352fa551159c5165a6a77718ae4ade316aedfea4b1c2
Windows x64 CUDA	funasr-llamacpp-windows-x64-cuda.zip	148657911fb666b7af6ec43af2e23a0984e3259012b4c39f95631b717feb6840
Windows x64 Vulkan	funasr-llamacpp-windows-x64-vulkan.zip	debf8007e55011cad06081e7b8a78972f1b8fe672bc324d41e650d68821f6a6a
Windows x64 portable	funasr-llamacpp-windows-x64.zip	f6a73a548413ba9fbaf2145263ea66ec53cbdad1fb11790dbeeee493e339492e

Quick start: download one asset, unpack it, then run the bundled download-funasr-model.sh <sensevoice|paraformer|nano> helper and one of llama-funasr-cli, llama-funasr-sensevoice, or llama-funasr-paraformer.

For Python users, install from PyPI:

python -m pip install -U "funasr==1.4.26"