stt is a speech-to-text tool that runs locally offline to convert voice in audio and video files to text.
The tool solves the problem of transcribing speech without relying on cloud APIs by using the fast-whisper open-source model. It processes audio and video files to recognize human speech and output results in multiple formats: JSON, SRT subtitles with timestamps, or plain text. The approach allows users to deploy it locally as a replacement for services like OpenAI's speech recognition API or Baidu speech recognition, with accuracy comparable to official APIs. The tool includes a built-in tiny model, with options to download larger models (base, small, medium, large-v3) that offer progressively better recognition at the cost of increased computational requirements.
Developers should choose this tool if they need offline speech recognition without API dependencies and have control over their deployment environment. It suits projects requiring batch processing of audio and video files, subtitle generation, or integration via its HTTP API. The tool provides both a web interface for manual uploads and a programmatic API endpoint for automation. Windows users can run a precompiled executable without setup, while Linux and macOS users deploy from source. The project automatically uses CUDA acceleration when an NVIDIA GPU with proper CUDA configuration is available.
Development activity shows consistent maintenance with regular updates addressing user needs. The project maintains documentation in both Chinese and English to serve its user base. The tool includes comprehensive deployment instructions for multiple operating systems and provides clear API documentation with support for numerous languages. The maintainers actively manage the model ecosystem by hosting downloadable model packages and offering flexible configuration options for different hardware capabilities.
