Parler-TTS
An open-source text-to-speech Python library that generates speech with voice styles described in text and supports training your own models.
Parler-TTS is a text-to-speech library for developers adding speech to applications and researchers adapting speech models. It generates audio with a voice style you describe in text.
- Voice descriptions: Control characteristics such as gender, pitch and speaking style through a text prompt.
- Consistent speakers: Choose a named speaker to keep the voice consistent across separate generations.
- Model choice: Parler-TTS Mini and Parler-TTS Large offer different model sizes, both trained on audiobook recordings.
- Custom training: Train or fine-tune a model on your own speech data, including data from a single speaker.
- Streaming generation: Generate speech as a stream, with support for SDPA, Flash Attention 2 and model compilation to improve inference speed.
The Python library supports local inference and Apple Silicon. The project is open source under Apache 2.0 and publishes model weights, datasets, preprocessing code and training code.
Logotyper
