ParlerTTS is a training and inference library for high-quality text-to-speech (TTS) models. This demonstration highlights the flexibility of the IndicParlerTTS model, which generates natural, expressive speech for over 22 Indian languages, using a simple text prompt to control features like speaker style, tone, pitch, pace, and more.

Tips for effective usage:

• Use detailed captions to describe the speaker and desired characteristics (e.g., "Aditi speaks in a slightly expressive tone, with clear audio quality and a moderate pace.").
• For best results, reference specific named speakers provided in the model card on the model page.
• Include terms like "very clear audio" or "slightly noisy audio" to control the audio quality and background ambiance.
• Punctuation can be used to shape prosody (e.g., commas add pauses for natural phrasing).
• If unsure about what caption to use, you can start with: "The speaker speaks naturally. The recording is very high quality with no background noise."

If you'd like to learn more about how the model was trained or explore fine-tuning it yourself, visit the Parler-TTS repository on GitHub. The Parler-TTS codebase and associated checkpoints are licensed under the Apache 2.0 license.