echogarden-project/echogarden
Cross-platform speech toolset, used from the command-line or as a Node.js library. Includes a variety of engines for speech synthesis, speech recognition, forced alignment, speech translation, voice isolation, language detection and more.
What it solves
Echogarden is a comprehensive speech processing toolset that simplifies the deployment of complex audio AI tasks. It removes the need for Python, Docker, or system-level dependencies by providing a unified TypeScript/Node.js environment for speech-to-text, text-to-speech, and audio analysis.
How it works
The project uses a combination of pure TypeScript implementations, WebAssembly ports, and the ONNX runtime to run AI models offline. It integrates a wide array of engines, including Kokoro and VITS for synthesis, OpenAI Whisper for recognition and translation, and MDX-NET for source separation. It also includes a system for automatic resource downloading and installation of models and voices.
Who it’s for
It is designed for developers who want to integrate speech capabilities into Node.js applications or users who prefer a command-line interface for audio processing without managing complex AI environments.
Highlights
- Multimodal Speech Tools: Supports text-to-speech (TTS), speech-to-text (STT), and speech-to-transcript alignment.
- Audio Enhancement: Includes voice activity detection (VAD), speech denoising, and source separation to isolate voices from background noise.
- Cross-Platform: Runs on Windows, macOS, and Linux (x64 and ARM64) without requiring Python.
- Advanced Translation: Offers speech-to-text translation and synchronization of spoken audio to translated transcripts.
- Pronunciation Improvements: Enhances TTS accuracy through text normalization and heteronym disambiguation for specific engines.