Skip to main content

​
Introduction to react-native-sherpa-onnx

A React Native TurboModule that provides offline and streaming speech processing capabilities using sherpa-onnx. Process speech entirely on-device with no internet connection required.

​
What is react-native-sherpa-onnx?

react-native-sherpa-onnx brings powerful speech processing capabilities to React Native applications:
  • Speech-to-Text (STT) - Convert audio to text offline
  • Text-to-Speech (TTS) - Generate natural-sounding speech from text
  • Streaming Recognition - Real-time speech recognition with partial results
  • Streaming TTS - Low-latency incremental speech generation
  • 100% Offline - All processing happens on-device with no internet required

​
Key Features

​
Offline Speech-to-Text

Transcribe audio files or samples without an internet connection. Supports multiple model architectures:
  • Zipformer/Transducer - Balanced speed and accuracy
  • Whisper - Multilingual, zero-shot capabilities
  • Paraformer - Fast non-autoregressive ASR
  • NeMo CTC - Excellent for English and streaming
  • SenseVoice - Emotion and punctuation detection
  • Moonshine - Lightweight streaming-capable models
  • And many more (see Supported Models)

​
Online (Streaming) Speech-to-Text

Real-time recognition from microphone or audio streams:
  • Partial results as the user speaks
  • Endpoint detection for natural pauses
  • Low latency for responsive UX
  • Use streaming-capable models (transducer, paraformer, nemo_ctc, tone_ctc)

​
Text-to-Speech

Generate high-quality speech from text:
  • VITS - Fast, high-quality (Piper, Coqui, MeloTTS)
  • Matcha - High-quality acoustic model with vocoder
  • Kokoro - Multi-speaker, multi-language
  • KittenTTS - Lightweight multi-speaker
  • Zipvoice - Voice cloning support

​
Streaming Text-to-Speech

Incremental speech generation for low time-to-first-byte:
  • Start playback while generating
  • Ideal for long texts
  • Chunk-based callbacks for streaming audio

​
Hardware Acceleration

Optimize performance with execution providers:
  • Android: CPU, NNAPI, XNNPACK, QNN (Qualcomm)
  • iOS: CPU, Core ML, Apple Neural Engine
  • Automatic detection and support checking

​
Flexible Model Loading

  • Asset models - Bundle models in your app
  • File system models - Download and use external models
  • Play Asset Delivery (PAD) - Android on-demand model delivery
  • Automatic detection - Auto-detect model types

​
Developer Experience

TypeScript Support

Full type definitions for all APIs

Instance-Based API

Multiple STT/TTS engines in parallel

Model Quantization

Automatic int8 model detection

Cross-Platform

iOS and Android production ready

​
Supported Models

​
Speech-to-Text Models

Model TypeUse CaseStreaming Support
Zipformer/TransducerBalanced speed/accuracy✅ Yes
WhisperMultilingual, zero-shot❌ Offline only
ParaformerFast inference✅ Yes
NeMo CTCEnglish, streaming✅ Yes
SenseVoiceEmotion detection❌ Offline only
MoonshineLightweight streaming✅ Yes
Tone CTC (t-one)Lightweight CTC✅ Yes
See the complete list in Model Types.

​
Text-to-Speech Models

Model TypeDescription
VITSFast, high-quality (Piper, Coqui, MeloTTS)
MatchaAcoustic model + vocoder
KokoroMulti-speaker, multi-language
KittenTTSLightweight multi-speaker
ZipvoiceVoice cloning with encoder/decoder
PocketFlow-matching TTS

​
Platform Support

PlatformStatusNotes
Android✅ Production ReadyAPI 24+ (Android 7.0+)
iOS✅ Production ReadyiOS 13.0+

​
Requirements

  • React Native >= 0.70
  • Android API 24+ (Android 7.0+)
  • iOS 13.0+
  • @dr.pogodin/react-native-fs (peer dependency for file operations)

​
Architecture

react-native-sherpa-onnx uses React Native TurboModules for high-performance native integration: 
┌─────────────────────────────────────┐
│   React Native JavaScript Layer     │
│  (TypeScript API + Type Safety)     │
└──────────────┬──────────────────────┘
               │ TurboModule Bridge
┌──────────────▼──────────────────────┐
│      Native Module (Obj-C/Kotlin)   │
│   Instance Management + Threading   │
└──────────────┬──────────────────────┘
               │ C++ API
┌──────────────▼──────────────────────┐
│         sherpa-onnx (C++)           │
│    ONNX Runtime + Model Inference   │
└─────────────────────────────────────┘

​
Why Choose react-native-sherpa-onnx?

All processing happens on-device. No data leaves the user’s phone, and no internet connection is required. Perfect for privacy-sensitive applications.
No API calls, no per-request costs, no rate limits. Once the model is bundled or downloaded, transcription and synthesis are completely free.
Direct on-device processing means no network round-trips. Streaming STT provides partial results in real-time, and streaming TTS can start playback within milliseconds.
Battle-tested in production apps with CI/CD automation, comprehensive documentation, and active maintenance.

​
Example Use Cases

  • Voice assistants - Offline voice commands and responses
  • Transcription apps - Convert meetings, lectures, or interviews to text
  • Accessibility tools - Text-to-speech for visually impaired users
  • Language learning - Real-time pronunciation feedback
  • Voice notes - Convert voice memos to searchable text
  • Healthcare apps - Medical transcription with privacy compliance
  • Navigation apps - Turn-by-turn voice guidance

​
What’s Next?

Installation

Install the library and set up iOS/Android

Quick Start

Get up and running with your first example

STT Guide

Learn about speech-to-text features

TTS Guide

Explore text-to-speech capabilities
Breaking changes in v0.3.0: If you’re upgrading from 0.2.x, see the Migration Guide for important API changes.

Build docs developers (and LLMs) love

Get started for freeTalk to us