Parler-TTS is Hugging Face’s open-source library for inference and training of high-quality text-to-speech models. It implements the Parler-TTS model family, which generates speech guided by a natural-language description: you describe the voice, speaking style and recording conditions in words, and the model follows. The library has more than 5,500 stars on GitHub.
Traditional TTS systems control voices through speaker IDs or reference audio. Parler-TTS instead treats voice attributes, such as gender, pitch, pace, emotion and background noise, as a text prompt, which makes expressive speech generation steerable in plain language. The repository provides everything needed to use and reproduce the approach: inference code for the released Large and Mini models, full training recipes, and data-processing tools, all integrated with the Hugging Face Transformers ecosystem and the Hub. Because the description prompt is separate from the text being spoken, the same sentence can be rendered as a calm studio narration, an energetic presentation or a noisy phone call simply by rewriting the description. That makes the library popular for generating training data, prototyping voice products and studying how attributes of speech can be controlled through language. Sample descriptions and generated audio in the project documentation illustrate the range of voices it can produce.
Free. Parler-TTS is open source under the Apache-2.0 license, including the released models; you only need hardware to run or train them.
Developers and researchers building voice interfaces, audiobook and content narration, or custom TTS systems who want controllable, expressive speech without a commercial API.