Starstruk Logo MarkStarstruk

v1.13.4

n4ze3m/page-assistv1.13.4Jul 7, 2026by csukuangfj

AI Summary

This release focuses on expanding hardware acceleration support by adding Qualcomm NPU (QNN) support for various speech recognition models like Parakeet, Whisper, and Nemotron, alongside C++ runtimes for these new exports.

Key Highlights

  • Added QNN support for Parakeet CTC, Whisper, and Nemotron-speech models
  • Introduced C++ runtimes for QNN-based speech recognition
  • Fixed Windows path handling issues using wide characters and C++17 std::filesystem
  • Updated ONNX Runtime to version 1.27.0
  • Added per-stream hotwords support in the JavaScript API

New Features

  • Export Parakeet CTC models to QNN
  • Export Whisper models to QNN
  • Add C++ runtime for Parakeet TDT models with QNN
  • Fix file related operations on Windows using wide characters
  • Support per-stream hotwords in JavaScript API

Full Release Notes

## What's Changed
* Fix releasing Dart iOS package by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3680
* Export Parakeet CTC models to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3686
* Add C++ runtime for Parakeet CTC with QNN. by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3688
* Build sdist in split-package mode so its metadata declares sherpa-onnx-core by @Eridanus117 in https://github.com/k2-fsa/sherpa-onnx/pull/3690
* Export Parakeet-TDT-CTC to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3692
* Export the transducer branch of parakeet-tdt-ctc to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3693
* Export Whisper models to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3697
* add project: Add SmartSub application details to README.md by @buxuku in https://github.com/k2-fsa/sherpa-onnx/pull/3698
* Add C++ runtime for Whisper with Qualcomm NPU using QNN. by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3699
* Fix building with ninja by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3702
* Fix #3171 : 去掉生成绝对路径函数中的窄字符WindowsAPI调用，改用宽字符API和C++17 std::filesystem来避免win平台中文路径问题 by @XiaoBaiCanDo in https://github.com/k2-fsa/sherpa-onnx/pull/3255
* Fix installing espeak-ng-data directory by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3708
* Remove spaces right before a punctuation in ASR result by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3709
* Fix file related operations on Windows by using wide characters by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3710
* Remove leading spaces from ASR results by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3711
* Update onnxruntime to 1.27.0 by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3718
* Export Parakeet TDT models to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3719
* Add C++ runtime for Parakeet TDT models with QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3720
* Export nemotron-speech-streaming-en-0.6b to QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3725
* Add C++ runtime for nemotron-speech-streaming-en-0.6b with QNN by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3728
* Add ApplyHomophoneReplacer in OfflineRecognizerQwen3ASRImpl by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3729
* Support per-stream hotwords in the JavaScript (node-addon) API for non-streaming ASR by @cmurrayis in https://github.com/k2-fsa/sherpa-onnx/pull/3723
* Release v1.13.4 by @csukuangfj in https://github.com/k2-fsa/sherpa-onnx/pull/3730

## New Contributors
* @Eridanus117 made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/3690
* @buxuku made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/3698
* @XiaoBaiCanDo made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/3255
* @cmurrayis made their first contribution in https://github.com/k2-fsa/sherpa-onnx/pull/3723

**Full Changelog**: https://github.com/k2-fsa/sherpa-onnx/compare/v1.13.3...v1.13.4