NeuTTS

NeuTTS is a collection of on-device text-to-speech models for developers building voice assistants and speech-enabled apps. It generates audio on your own hardware rather than sending speech generation to a cloud API.

• Voice cloning: NeuTTS-Air and NeuTTS-Nano can reproduce a speaker's voice from a clean audio sample as short as three seconds and its matching transcript.
• Language-specific models: Air handles English, while the Nano collection has separate models for English, French, German and Spanish; references in the target language give better results.
• Emotional speech: NeuTTS-2E generates English speech with four fixed speakers and selectable emotions, including happiness, sadness and anger, plus a neutral setting.
• Streaming audio: GGUF models can play speech as it's generated, so an app doesn't have to wait for the complete audio clip.

The Python toolkit runs on macOS, Windows and Linux. Models come in PyTorch and quantized GGUF formats, with llama-cpp-python support for GGUF inference. It supports CPU operation and hardware acceleration with NVIDIA, AMD and Apple Silicon GPUs; the models also target phones and Raspberry Pis.

NeuTTS-Air uses Apache 2.0. Nano and 2E use the custom NeuTTS Open License 1.0. The context accommodates roughly 30 seconds of audio, including the reference prompt. Generated audio includes a Perth watermark by default when the watermarking dependency is available.