StreamASR is a high-performance real-time speech recognition service that provides WebSocket interface for real-time audio stream transcription (converting OpenAI standard v1/audio/transcriptions interface to real-time speech recognition service). The project integrates VAD (Voice Activity Detection) and multiple ASR engines, supporting flexible configuration and deployment.
- 🎤 Real-time Speech Recognition - Low-latency audio stream processing based on WebSocket
- 🧠 Smart VAD Detection - Integrated Sherpa-ONNX voice activity detection with automatic audio submission trigger
- 🔇 AI Noise Reduction - Built-in denoiser using GTCRN model for enhanced speech recognition in noisy environments
- 🔄 OpenAI Compatible - Supports OpenAI-compatible ASR interface with configurable multiple models
- 📊 Structured Logging - Detailed logging and monitoring based on logrus
- 🐳 Docker Support - Complete containerized deployment solution
- 🔧 Version Management - Automated version management and build process
- 🌐 Multi-language SDK - Provides Go and TypeScript client SDKs
- Go 1.23+ - Server runtime environment
- VAD Model File - Sherpa-ONNX VAD model (silero_vad.onnx)
- Denoiser Model File - Sherpa-ONNX GTCRN model (gtcrn_simple.onnx)
- ASR Service - OpenAI-compatible speech recognition API
# Clone the project
git clone https://github.com/go-restream/stt.git
cd stt
# Install dependencies and build
make install
make build
# Start the service
make run# Install dependencies
go mod download
# Build the project
go build -o streamASR main.go
# Start the service
./streamASR -c config.yaml# Using docker-compose
make docker-deploy
# Or manual build
make docker-build
make docker-compose-upAfter the service starts, you can verify it through:
# View version information
./build/streamASR -v
# Health check
curl http://localhost:8088/health
# Check service status
curl http://localhost:8088/statusStreamASR provides a built-in Web UI tool that allows users to perform real-time speech recognition testing directly through their browser.
After starting the service, visit in your browser:
# Main interface
http://localhost:8088/
# Or directly access static files
http://localhost:8088/static/index.html- 🎤 Real-time Audio Visualization - Dynamic display of audio waveforms and volume levels
- 🔧 Configuration Options - Support for sample rate selection (16kHz/48kHz) and VAD toggle
- ⚡ Real-time Transcription - Real-time display of speech recognition results
- 🎨 Theme Switching - Support for multiple visual themes (Deep Blue Tech, Purple Cyber, Green Matrix)
- 💾 Result Saving - Support for copying and saving transcription results
- 🤖 AI Summary - Integrated AI functionality for intelligent summarization of transcription content
- Open Browser Visit
http://localhost:8088 - Configure Parameters Select sample rate and VAD detection toggle
- Click Start Launch speech recognition
- Authorize Microphone Browser will request microphone permission
- Start Speaking View real-time transcription results
- Save Results Use save button to copy transcription text
- WebSocket Connection - Low-latency communication based on WebSocket
- Auto Reconnection - Support for automatic reconnection on disconnection
- Heartbeat Detection - 30-second heartbeat to maintain stable connection
- Error Handling - Comprehensive error prompts and status display
# Service port configuration
service_port: "8088"
# OpenAI compatible ASR interface configuration
asr:
base_url: "http://localhost:3000/v1" # ASR interface base URL
api_key: "your-api-key" # ASR interface API key
model: "FireRed-large" # ASR model name
# OpenAI compatible LLM interface configuration (optional)
llm:
base_url: "https://api.deepseek.com/v1" # LLM interface base URL
api_key: "your-llm-api-key" # LLM interface API key
model: "deepseek-chat"
# Audio configuration
audio:
enable: true
save_dir: "./audio" # Audio file save directory
keep_files: 10 # Keep recent wav file records
sample_rate: 16000 # Sample rate (16kHz/48kHz)
channels: 1 # Number of channels
bit_depth: 16 # Bit depth
buffer_size: 10 # 10-second buffer
# VAD configuration
vad:
enable: true
model: "./model/silero_vad.onnx" # VAD model path
threshold: 0.5 # Speech detection threshold
min_silence_duration: 1 # Minimum silence duration (seconds)
min_speech_duration: 0.1 # Minimum speech duration (seconds)
window_size: 512 # Window size
max_speech_duration: 8.0 # Maximum speech duration (seconds)
sample_rate: 16000 # Sample rate
num_threads: 1 # Number of threads
provider: "cpu" # Compute provider
# Denoiser configuration (AI Noise Reduction)
denoiser:
enable: true # Enable/disable denoiser
model: "./model/gtcrn_simple.onnx" # GTCRN denoiser model path
sample_rate: 16000 # Sample rate
num_threads: 1 # Number of threads
debug: 0 # Debug level (0-3)
bypass_for_testing: false # Bypass denoiser for testing
max_processing_time_ms: 50 # Maximum processing time (ms)
# Logging configuration
logging:
level: "info" # Log level
file: "" # Log file path, empty means output to stderr
format: "json" # Log format: json, text# One-command deployment (build and start all services)
make docker-deploy
# Check service status
make docker-ps
# View real-time logs
make docker-compose-logs
# Stop all services
make docker-compose-downThe docker-compose.yml provides complete service orchestration:
version: '3.8'
services:
streamASR:
build: .
ports:
- "8088:8088"
volumes:
- ./config/config.yaml:/app/config/config.yaml:ro
- ./build/model:/app/model:ro
- ./audio:/app/audio
- ./logs:/app/logs
environment:
- VERSION=v0.1.2
restart: unless-stopped
healthcheck:
test: ["CMD", "curl", "-f", "http://localhost:8088/health"]
interval: 30s
timeout: 10s
retries: 3# Build Docker image
make docker-build
# Run container with volume mounts
make docker-run
# Enter container for debugging
make docker-exec
# View container logs
make docker-logs
# Stop and remove container
make docker-stop# Build development image and run interactive container
make docker-debug
# Run in development mode with hot reload
make docker-dev- Multi-stage build - Optimizes image size
- Version information injection - Automatically injects version, build time and other information
- Health check - Built-in health check mechanism
- Non-root user - Secure container runtime environment
- Production-ready - Optimized for production deployment
# List all containers
docker ps -a
# Monitor resource usage
docker stats
# Clean up unused resources
make docker-cleanFor detailed Docker deployment guide, please refer to: docs/DOCKER.md | English Version
# Display version information
./streamASR -v
./streamASR --version
# Specify configuration file
./streamASR -c config.yaml
# View help information
./streamASR -hpackage main
import (
"streamASR/sdk/golang/client"
)
func main() {
// Create client
recognizer := client.NewRecognizer("ws://localhost:8088")
// Connect and start recognition
err := recognizer.Connect()
if err != nil {
panic(err)
}
defer recognizer.Close()
// Handle audio...
}import { StreamASRClient } from '@streamasr/typescript-sdk';
const client = new StreamASRClient({
url: 'ws://localhost:8088',
autoConnect: true
});
// Listen for transcription results
client.on('transcription', (result) => {
console.log('Recognition result:', result.text);
});The service uses logrus for structured logging:
{
"component": "mont_srv_status",
"action": "health_check_status",
"version": "v0.1.2-171f62c",
"build_time": "2025-11-02T05:24:39Z",
"git_commit": "171f62c",
"level": "info",
"msg": "✔ Starting StreamASR v0.1.2-171f62c with config: config.yaml"
}# Basic health check
curl http://localhost:8088/health
# Return example
{
"status": "healthy",
"version": "v0.1.2-171f62c",
"uptime": "2h30m15s",
"asr_engine": "available"
}The project adopts semantic version management and supports automated version releases:
# View current version
make version
# Version upgrade
make version-bump-patch # v0.1.2 -> v0.1.3
make version-bump-minor # v0.1.2 -> v0.2.0
make version-bump-major # v0.1.2 -> v1.0.0
# Create Git tag
make tag
# Build Docker image
make docker-build # Generate streamasr:latest and streamasr:v0.1.2For detailed version management guide, please refer to: docs/VERSION.md
# Clone project
git clone https://github.com/go-restream/stt.git
cd stt
# Install dependencies
make install
# Run tests
make test
# Build
make build
# Run
make run# Docker development mode
make docker-debug
# View logs
make docker-logs
# Enter container for debugging
make docker-exec# Run unit tests
make test
# Run integration tests
go test ./...-
VAD Model File Missing
# Ensure VAD model file exists ls -la model/silero_vad.onnx -
Denoiser Model File Missing
# Ensure denoiser model file exists ls -la model/gtcrn_simple.onnx -
ASR Service Connection Failed
# Check ASR service configuration curl -H "Authorization: Bearer $API_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"FireRed-large","file":"..."}' \ $ASR_BASE_URL/audio/transcriptions
-
Port Occupied
# Check port occupation lsof -i :8088 # Modify port in configuration file vim config.yaml
Enable verbose logging:
# Modify configuration file
vim config.yaml
# Set logging.level: "debug"
# Or set environment variable
export LOG_LEVEL=debug
./streamASR- Response Latency: < 500ms end-to-end recognition latency
- Concurrency Support: Supports multiple concurrent WebSocket connections
- Audio Processing: Supports 16kHz/48kHz sample rates
- VAD Latency: < 100ms voice activity detection latency
- Denoiser Latency: < 20ms additional processing time for noise reduction
- Noise Reduction: Improved ASR accuracy in noisy environments
We welcome community contributions! Please follow these steps:
- Fork the project repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Commit changes (
git commit -m 'Add amazing feature') - Push to the branch (
git push origin feature/amazing-feature) - Create a Pull Request
- Follow Go coding standards
- Add unit tests
- Update relevant documentation
- Pass all CI checks
- Docker Deployment Guide - Complete Docker deployment instructions
- Version Management Documentation - Version management specifications and usage
- Changelog - Detailed version change records
| Method | Description | Response Time |
|---|---|---|
| GitHub Issues | Bug reports and feature requests | 24-48 hours |
| GitHub Discussions | Technical discussions and Q&A | Community response |
- 🔇 AI Noise Reduction - Built-in denoiser using GTCRN model for enhanced speech recognition
- 🏷️ Version Management System - Complete version management and build process
- 🐳 Docker Support - Complete containerized deployment solution
- 📋 Makefile Integration - Automated build and deployment scripts
- 📖 Documentation Enhancement - Detailed deployment and development documentation
- 🎯 Audio Pipeline Enhancement - Integrated denoiser between VAD and ASR processing
- 🔧 Project Structure Optimization - Clearer code organization and module division
- 📝 Logging Enhancement - Startup logs include version information
- 🛠️ Build Process - Support for automatic version information injection
- 🧪 Comprehensive Testing - Unit tests for denoiser functionality
- 🎤 Real-time Speech Recognition - WebSocket-based audio stream processing
- 🧠 VAD Integration - Sherpa-ONNX voice activity detection
- 🔄 ASR Interface - OpenAI-compatible speech recognition API
- 📊 Health Check - Service status monitoring interface
StreamASR is a feature-complete, easy-to-deploy real-time speech recognition service. Through Docker containerization, version management system, and comprehensive documentation, it provides a reliable speech recognition solution for production environments.
⭐ If this project helps you, please give us a Star!
🎯 StreamASR - Making Speech Recognition Simple and Powerful