sherpa-onnx
Local speech-to-text, text-to-speech, and audio processing offline toolkit.
Run speech recognition, diarization, and VAD locally without internet connection.
sherpa-onnx provides offline speech-to-text, text-to-speech, speaker diarization, and VAD powered by next-gen Kaldi and onnxruntime. It is built for mental health professionals and developers who need to process therapy session audio entirely on local devices without exposing sensitive data to cloud services.
What it is
The tool performs local speech recognition, speaker identification, diarization, audio tagging, source separation, and voice activity detection. It runs completely offline on embedded systems, mobile devices, desktops, and servers across 12 different programming languages.
Who it's for
It suits clinicians, researchers, and developers building secure, offline-first applications for analyzing or transcribing therapy sessions. It does not suit practitioners looking for a turnkey, cloud-hosted SaaS practice management platform with zero technical setup required.
Key features
Offline Speech Recognition
Transcribe audio locally using streaming or non-streaming models without internet access.
Speaker Diarization
Identify and separate different speakers in a multi-person conversation like therapy sessions.
Voice Activity Detection
Detect speech segments accurately using tools like silero-vad embedded locally.
Multi-Platform Support
Deploy across Android, iOS, Linux, Windows, macOS, Raspberry Pi, and specialized NPUs.
Up and running in 3 steps
Our honest verdict
What's great
- Runs entirely offline ensuring complete data privacy for sensitive therapy audio
- Supports 12 programming languages including Python, C++, Java, and Swift
- Provides pre-built demo apps for Flutter and Tauri
- Includes speech enhancement, source separation, and punctuation restoration
Where it falls short
- Requires significant technical skill to compile, integrate, and deploy across target platforms
- Does not include a ready-to-use graphical user interface tailored specifically for mental health workflows
- You must source, select, and manage your own speech and AI models locally
Pricing & license
Free under its Apache-2.0 licence, with no paid hosted options provided by the project. The only cost is running it yourself on your own hardware.
Open remote AI-training roles
876 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Does sherpa-onnx require an internet connection to run?
No, all speech recognition, text-to-speech, and audio processing tasks run completely offline.
What programming languages are supported?
It supports 12 languages including C++, C, Python, JavaScript, Java, C#, Kotlin, Swift, Go, Dart, Rust, and Pascal.
Can I run this on mobile devices?
Yes, it supports Android, iOS, and HarmonyOS with pre-built Flutter and Tauri demo apps available.
