drama3-expressive-medium

Built with Fish Audio. Japanese / English speech that follows acting directions, including ones placed in the middle of a line, with drama3-expressive-medium, a fine-tuned and quantized derivative of Fish Audio Drama 3. This demo runs the model's PyTorch runtime from the GGUF. Fish Audio does not sponsor or endorse it.

Put a direction in square brackets at the start of the line, and another right before the word where the delivery should change: [cheerful] Good morning! [suddenly worried] Wait, where's my phone? Write directions in English for Japanese lines too; 2–8 words work best. Lead with the feeling (angry, heartbroken, amused, nervous, tender), then how it is voiced (whispering, voice breaking, through laughter, rushed).
Examples
Voice
Voice ideas

No reference audio: the model makes the voice up, in the language of the text. Every seed is a different voice. Like one? Press Keep this voice under the result to use it for more lines. Describe a kind of voice, not a real person.
0.3 1.2
0.3 1
1 100
0.3 1
5 40

Terms of use. This is a free, non-commercial research demo. By generating audio you agree that you will not use this demo or its output to:

  • imitate the voice of an identifiable person without that person's prior consent;
  • impersonate any person or organization to deceive, defraud or harass, or place calls that break telemarketing or robocall laws;
  • create sexual content involving a minor or a voice presented as a minor's, or using a real person's voice without consent;
  • remove, disable or obscure any watermark or provenance signal, or train, distill or improve other speech or voice generation models.

Only upload voices that are your own or that you have the speaker's explicit consent to clone. Uploaded clips are used for that request only and are not stored. Output is synthetic speech and provided as is.

drama3-expressive-medium is a derivative of Fish Audio Drama 3, licensed under the Fish Audio Community License v1.0 (see LICENSE and NOTICE). Generated voices come from the model alone, with no reference audio. Takes get 0.1 s of silence at each end.