Starstruk Logo MarkStarstruk

v1.12.19

k2-fsa/sherpa-onnxv1.12.19Dec 5, 2025by csukuangfj

AI Summary

Version 1.12.19 focuses on stability and cross-platform support, fixing bugs in ZipVoice tokenization, NeMo speaker embeddings, and Matcha TTS. It adds support for Spacemit RISC-V, AXERA NPUs, and token-level confidence scores for transducers.

Key Highlights

  • Added support for Spacemit RISC-V CPU (ort ep).
  • Added support for AXERA NPU backends (ax630, ax650, axcl).
  • Added token-level confidence scores (ys_probs) for offline transducer models.
  • Fixed ZipVoice English tokenization errors.
  • Fixed NeMo speaker embedding NaN values.

New Features

  • Spacemit ORT EP support
  • AXERA NPU support
  • Token-level confidence scores
  • ZipVoice fixes
  • Matcha TTS fixes

Full Release Notes

## What's Changed
* Fix building without TTS for C API by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2838
* [ZipVoice] Fix english tokenization error by @pkufool in https://github.com/k2-fsa/sherpa-onnx/pull/2834
* Add simulate streaming ASR Python example for Paraformer by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2839
* Fix building JNI for Windows. by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2840
* Avoid NaN in NeMo speaker embedding models. by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2844
* add spacemit ort ep for spacemit riscv cpus by @alex-spacemit in https://github.com/k2-fsa/sherpa-onnx/pull/2837
* Add token-level confidence scores (ys_probs) for offline transducer models by @Mahmoud-ghareeb in https://github.com/k2-fsa/sherpa-onnx/pull/2843
* Fix token log probabilities in offline transducer modified beam search decoder by @Mahmoud-ghareeb in https://github.com/k2-fsa/sherpa-onnx/pull/2846
* Support AXERA ax630, ax650, and axcl backends. by @Abandon-ht in https://github.com/k2-fsa/sherpa-onnx/pull/2849
* Refactor axera npu examples by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2850
* fix matcha tts zh-en model by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2851
* Fix the English part for Matcha TTS. by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2853
* Refactor text-utils by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2855
* Fix matcha tts by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2856
* Add a space between English words for Matcha zh-en TTS by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2858
* Fix punctuations in matcha zh-en tts by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2859
* Upload matcha tts zh-en model by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2865
* Fix the discrepancy with the Silero VAD isSpeech logic by @ming030890 in https://github.com/k2-fsa/sherpa-onnx/pull/2863
* Release v1.12.19 by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/2868

## New Contributors
* @alex-spacemit made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/2837
* @Mahmoud-ghareeb made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/2843
* @Abandon-ht made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/2849
* @ming030890 made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/2863

**Full Changelog**: https://github.com/k2-fsa/sherpa-onnx/compare/v1.12.18...v1.12.19