Supports: http://, https://, or hf:// URLs.
You can also use predefined voices:
"alba", "anna", "azelma", "bill_boerst", "caro_davy", "charles", "eponine", "eve",
"fantine", "george", "jane", "jean", "mary", "michael", "paul", "peter_yearsley", "stuart_bell", "vera".
You can find more voices in our
voice repository.
Upload an audio file (WAV, MP3, FLAC, etc.) to use as voice reference. Takes precedence over voice URL. For best results, clean the sample first.
First use of a new language takes extra time to download the model (~few minutes).
Higher = more expressive/varied, lower = more stable/flat.
Controls when speech stops. Lower = waits longer before ending.