speak-rs

A powerful desktop application for real-time speech transcription using Rust and Whisper AI. This application provides high-quality speech-to-text conversion with GPU acceleration support and a modern user interface.

Features

• Real-time Transcription: Convert speech to text in real-time using your system's microphone
• GPU Acceleration: CUDA support for faster transcription processing
• Automatic Clipboard Integration: Automatically copy transcribed text to clipboard
• Configurable Settings: Easy customization through TOML configuration
• Stop Phrase Detection: Automatically stop transcription when a specific phrase is detected
• Modern UI: Built with Slint for a native and responsive user experience
• Recording Duration Display: Real-time display of recording duration in minutes and seconds
• Transcription Status Indicator: Visual feedback when transcription is in progress
• Clean Interface: Simple record/stop button with clear transcription display
• Error Handling: Robust error handling with user-friendly error messages
• Resource Management: Automatic cleanup of resources on application exit

Prerequisites

• Rust (latest stable version)
• CUDA toolkit (for GPU acceleration)
• System with a microphone
• Linux with Wayland support (current implementation)

Installation

1. Clone the repository:

   git clone https://github.com/ThilinaTLM/speak-rs.git
   cd speak-rs

2. Download the Whisper model:

   • Create a models directory
   • Download the desired model (e.g., ggml-small.en.bin) into the models directory

3. Build the application:

   cargo build --release

Configuration

The application can be configured through config.toml:

[whisper]
model_path = "models/ggml-small.en.bin"       # Path to Whisper model
use_gpu = true                                # Enable GPU acceleration
language = "en"                               # Target language
audio_context = 768                           # Audio context size
no_speech_threshold = 0.5                     # Threshold for no speech detection
num_threads = 2                               # Number of CPU threads to use

[behavior]
realtime_transcribe = true                    # Enable real-time transcription
auto_copy = true                              # Automatically copy text to clipboard
stop_phrase_enabled = true                    # Enable stop phrase detection
stop_phrase_pattern = "(?i)that'?s all\\.?$"  # Regex pattern for stop phrase

Usage

1. Run the application:

   cargo run --release

2. The application will start with a GUI interface
3. Click the record button to start transcription
4. Speak into your microphone
5. The transcribed text will appear in real-time
6. Monitor recording duration in the interface
7. Use the configured stop phrase (default: "that's all") to automatically stop recording
8. Transcribed text is automatically copied to your clipboard if enabled
9. Click the record button again to stop manually, or use the close button to exit

Features in Detail

Real-time Transcription

• Continuous transcription during recording
• Updates every few seconds with new content
• Visual indicator when transcription is in progress

Recording Controls

• Simple record/stop toggle button
• Duration display showing minutes and seconds
• Automatic resource cleanup on stop

Text Processing

• Automatic removal of stop phrases from final text
• Smart handling of partial transcriptions
• Clipboard integration for easy text sharing

Error Handling

• Graceful handling of transcription errors
• User-friendly error messages
• Automatic recovery from common issues

Dependencies

• whisper-rs: Rust bindings for OpenAI's Whisper model
• cpal: Cross-platform audio library
• slint: Modern UI framework
• rubato: Audio resampling
• clap: Command line argument parsing
• config: Configuration management
• And more (see Cargo.toml for full list)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

• OpenAI for the Whisper model
• The Rust community for excellent libraries and tools