Voice Input Service

High-quality voice-to-text with advanced hallucination filtering and intelligent silence detection

A multi-platform voice input solution featuring superior transcription quality through advanced text processing and voice activity detection.

🌟 What Makes This Special

Unlike basic Whisper wrappers, this service includes:

• ✨ Advanced Hallucination Filtering - Removes common Whisper artifacts ("thanks for watching", timestamps, etc.)
• 🎯 Silero VAD Integration - State-of-the-art voice activity detection
• 🔄 Smart Text Processing - Intelligent overlap detection and text appending
• 🎤 Multiple Modes - Continuous (auto-pause) and manual recording modes
• 🌍 Multi-language Support - Works with all Whisper-supported languages

📱 Platforms

Desktop (Windows/macOS/Linux)

Real-time voice-to-text service with system-wide text insertion capabilities.

Status: ✅ Production Ready

• Desktop Documentation
• Installation Guide

Android

System-wide voice keyboard (IME) bringing desktop-quality filtering to mobile.

Status: ✅ IME Complete!

• Professional voice keyboard with dual modes (Tap/Hold)
• Real-time audio visualizer
• Advanced hallucination filtering
• Cosmos-themed modern UI
• See Current Status for details

🚀 Quick Start

Desktop

# Clone repository
git clone https://github.com/yourusername/STT
cd STT/desktop

# Install dependencies
python -m venv .venv
source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -e ".[dev]"

# Run
python -m voice_input_service

Keyboard Shortcuts:

• Alt+R - Start/Stop recording
• Alt+I - Toggle insertion mode
• Alt+V - Paste text
• Alt+S - Save transcript
• Alt+C - Clear text

Android

🚧 Android version is under active development. See Android Port Plan for progress.

📚 Documentation

For Users

• Desktop User Guide - Installation and usage
• Configuration Guide - Customizing settings
• Troubleshooting - Common issues

For Developers

• Architecture Overview - System design
• Core Algorithms - Shared logic documentation
• API Documentation - Technical details
• Development Guide - Contributing

Android Development

• Android Port Plan - Complete execution roadmap
• Quick Start - Jump in fast
• Component Mapping - Python to Kotlin

🏗️ Architecture

Core Components (Shared Logic)

The heart of our quality is in these platform-agnostic algorithms:

1. Text Processor (docs)

   • 40+ hallucination patterns
   • Intelligent overlap detection
   • Smart text appending with capitalization

2. Voice Activity Detection (docs)

   • Silero VAD with configurable threshold
   • Frame-based processing (30ms)
   • Device optimization (CPU/GPU/NNAPI)

3. Audio Processing Pipeline (docs)

   • Buffer management
   • Silence-triggered segmentation
   • Continuous and manual modes

Platform Implementations

Desktop (Python)

Microphone → AudioRecorder → TranscriptionWorker (VAD) →
Whisper → TextProcessor → System Text Insertion

Android (Kotlin - In Progress)

AudioRecord → AudioProcessor (VAD) → WhisperEngine →
TextProcessor → IME Text Insertion

🎯 Features

Text Quality

• ✅ Removes Whisper hallucinations automatically
• ✅ Filters timestamps and artifacts
• ✅ Intelligent text deduplication
• ✅ Proper capitalization and spacing
• ✅ Configurable minimum word count

Voice Detection

• ✅ Silero VAD (neural network-based)
• ✅ Configurable sensitivity (0.0-1.0)
• ✅ Smart silence duration detection
• ✅ Automatic vs manual modes

Transcription

• ✅ OpenAI Whisper (tiny → large models)
• ✅ whisper.cpp support (faster)
• ✅ Multi-language support
• ✅ GPU acceleration (CUDA/MPS/CPU)

User Experience

• ✅ System-wide operation
• ✅ Direct text insertion or clipboard
• ✅ Configurable keyboard shortcuts
• ✅ Session saving with timestamps
• ✅ Real-time transcription display

🔧 Configuration

Audio Settings

sample_rate: 16000          # Hz
vad_threshold: 0.5          # 0.0-1.0 (higher = less sensitive)
silence_duration_sec: 2.0   # Seconds of silence to trigger processing
max_chunk_duration_sec: 15.0 # Maximum audio chunk size

Transcription Settings

model_name: "base"          # tiny, base, small, medium, large
language: "en"              # Language code or None for auto-detect
use_cpp: true               # Use whisper.cpp (faster)

Full configuration guide: Configuration Documentation

📊 Quality Comparison

Feature	Basic Whisper	Other Apps	Voice Input Service
Hallucination Filtering	❌	Basic	✅ Advanced (40+ patterns)
Overlap Detection	❌	❌	✅ Smart deduplication
VAD Quality	Basic	Varies	✅ Silero (SOTA)
Text Processing	None	Basic	✅ Intelligent appending
Continuous Mode	❌	Limited	✅ Fully featured

🤝 Contributing

We welcome contributions! This project is unique in its quality focus.

For Desktop

See Desktop Development Guide

For Android

See Android Port Plan - We're actively porting to Android!

Areas Needing Help

• 🟢 Easy: Testing on different devices/OS
• 🟡 Medium: Additional hallucination patterns
• 🔴 Advanced: Android IME development

📈 Roadmap

Completed ✅

• Desktop voice input service
• Advanced hallucination filtering
• Silero VAD integration
• whisper.cpp support
• Continuous mode with smart appending
• Comprehensive documentation

In Progress 🚧

• Android Settings & Configuration UI
• Main app polish
• Real-world testing in popular apps

Planned 📋

• iOS version
• Web app (browser-based)
• Cloud sync for transcripts
• Custom model fine-tuning
• Multi-speaker diarization

🐛 Known Issues

See GitHub Issues for current bugs and feature requests.

Desktop:

• FFmpeg required for Whisper
• Large models slow on CPU
• Windows-only keyboard shortcuts (for now)

Android:

• In development - not yet functional

📝 License

MIT License - See LICENSE file for details.

🙏 Acknowledgments

• OpenAI Whisper team for the excellent STT model
• ggerganov for whisper.cpp
• Silero team for VAD models
• All contributors and testers

💬 Support

• 📖 Documentation
• 🐛 Report Issues
• 💡 Feature Requests
• 📧 Email: [email protected]

---

Built with ❤️ for high-quality voice input

⭐ Star this repo if you find it useful!

---

Repository Structure

STT/
├── android/              # Android application (in development)
├── desktop/              # Desktop application (Python)
├── docs/                 # Documentation
│   ├── core/            # Shared algorithm docs
│   ├── desktop/         # Desktop-specific docs
│   └── android/         # Android-specific docs
├── ANDROID_PORT_PLAN.md # Detailed Android porting plan
├── QUICKSTART.md        # Quick start for contributors
└── README.md            # This file

For developers: Start with QUICKSTART.md to understand the project structure and begin contributing.