This is a simplified guide to an AI model called F5-Tts maintained by X-Lance. If you like these kinds of analysis, you should join AImodels.fyi or follow us on Twitter.
Model overview
F5-TTS is a state-of-the-art open-source voice cloning model developed by x-lance. It is an improvement over the previous E2-TTS model, featuring a Diffusion Transformer with ConvNeXt V2 that enables faster training and inference. The model uses a novel "Sway Sampling" inference-time flow step sampling strategy to greatly improve performance.
Similar open-source voice cloning models include OpenVoice v2, Tortoise TTS, XTTS-v2, Parler TTS, and WhisperSpeech Small, each with their own unique capabilities and use cases.
Model inputs and outputs
F5-TTS is a text-to-speech (TTS) model that can generate high-quality synthetic speech from input text. The model can also perform voice cloning, allowing users to generate speech that mimics the voice of a reference audio sample.
Inputs
- Gen Text: The text to generate speech for
- Ref Text: The reference text to align the generated speech with
- Ref Audio: The reference audio sample to clone the voice from
- Speed: The speed of the generated audio, ranging from 0.1 to 3
- Remove Silence: A boolean flag to automatically remove silences from the generated audio
- Custom Split Words: A comma-separated list of custom words to split the generated text on
Outputs
- Output: A URI to the generated speech audio
Capabilities
F5-TTS is capable of generating high...
Top comments (0)