supertone-inc/supertonic
Supertonic
Supertonic is an on-device neural text-to-speech engine that runs entirely locally without cloud dependencies or GPU acceleration. It converts written text into natural-sounding speech across 31 languages with automatic language detection and a fallback model for unsupported locales.
The engine provides expressive speech control through inline prosody tags that dynamically adjust pitch, rate, and tone during synthesis. It supports voice cloning from a short reference audio clip by extracting a speaker embedding vector, and offers a selection of pre-built voices tuned for different use cases. Synthesis quality and playback speed are configurable, allowing trade-offs between latency and audio fidelity.
Supertonic includes a local HTTP server that exposes an endpoint compatible with the OpenAI Audio Speech API specification for drop-in integration with external tools and applications. A command-line interface is also available for generating audio files directly from text with configurable voice, quality, language, and style parameters. The engine applies rule-based and ML-based text normalization for numbers, dates, currencies, and units across supported languages.
Features
- Command-Line Speech Synthesizers - Generates audio files from text directly in the terminal with options for voice selection, quality, and language.
- Compact Neural TTS Models - Ships a compact neural TTS model that runs entirely on-device without cloud or GPU dependencies.
- Text-to-Speech Conversions - Converts written text into natural-sounding speech using a compact on-device model supporting 31 languages.
- Multilingual Text-to-Speech Engines - Converts written text into natural-sounding speech across 31 languages with automatic language detection.
- CLI Speech Synthesizers - Generates audio files from text via a command-line interface with configurable voice, quality, and language.
- Multilingual Speech Synthesizers - Converts written text into natural-sounding speech across 31 languages with automatic language detection.
- On-Device Text-to-Speech Synthesizers - Provides a compact neural TTS model that runs entirely on-device without cloud dependencies or GPU acceleration.
- Expressive Prosody Controls - Applies inline tags to alter tone, pitch, or rhythm for natural human expression in generated audio.
- OpenAI-Compatible Audio Servers - Exposes a local HTTP server implementing the OpenAI Audio API specification for drop-in TTS integration.
- Inline-Tag Styling - Parses embedded tags in input text to dynamically adjust pitch, rate, and tone during synthesis.
- Prosody Tag Parsers - Parses inline prosody tags to dynamically adjust pitch, rate, and tone during speech synthesis.
- Zero-Shot Voice Cloning - Loads a voice style from a JSON file, enabling zero-shot voice cloning from a short reference clip.
- Voice Identity Selections - Offers 10 distinct male and female voices tuned for specific tonal qualities and use cases.
- Voice Cloning - Extracts speaker embeddings from short audio clips to clone voice characteristics without fine-tuning.
- Voice Embedding Precomputations - Extracts a speaker embedding vector from a short audio clip to clone voice characteristics without fine-tuning.
- Multilingual Normalization Pipelines - Applies rule-based and ML-based normalization for numbers, dates, currencies, and units across 31 languages.
- Speed-Quality Tradeoffs - Offers configurable quality tiers and playback speed parameters that trade off latency against audio fidelity.
- AI Tools - High-speed local TTS engine.
