Overview
Sherpa-ONNX is a Flutter/Dart package enabling offline speech recognition (ASR), speech synthesis (TTS), speaker diarization, and speaker recognition using next-generation Kaldi with ONNX Runtime. It operates entirely without internet connectivity, making it ideal for privacy-sensitive or low-latency applications. Built for cross-platform deployment, it supports Android, iOS, macOS, Windows, and Linux. The package leverages optimized ONNX models for efficient on-device inference, ensuring fast and accurate voice processing directly on the device.
Use cases
- Offline voice assistants
- Real-time transcription apps
- Speaker identification in meetings
- Privacy-focused audio analysis
- On-device TTS for accessibility
Key features
- Fully offline operation
- Cross-platform support
- Low-latency inference
- Supports ASR, TTS, diarization, and recognition
- Optimized with ONNX Runtime
Suitable for
- Developers building offline voice apps
- Applications requiring data privacy
- Edge computing scenarios
- Mobile and desktop voice tools
- Real-time audio processing
Considerations
- Large model file sizes
- Higher memory usage during inference
- Requires model files to be bundled
- Limited to supported model types
- Initial setup complexity for custom models