Seed Voice Conversion

Choose between V1 (Voice & Singing Voice Conversion) or V2 (Voice & Style Conversion)

Seed Voice Conversion V2 (Voice & Style Conversion)

Zero-shot voice/style conversion with in-context learning.for details and updates.
Note that any reference audio will be forcefully clipped to 25s if beyond this length.
If total duration of source and reference audio exceeds 30s, source audio will be processed in chunks.
Please click the 'convert style/emotion/accent' checkbox to convert the style, emotion, or accent of the source audio, or else only timbre conversion will be performed.
Click the 'anonymization only' checkbox will ignore reference audio but convert source to an 'average voice' determined by model itself.

Source Audio

Reference Audio

Diffusion Steps

30 by default, 50~100 for best quality

1 200

Length Adjust

<1.0 for speed-up speech, >1.0 for slow-down speech

0.5 2

Intelligibility CFG Rate

controls pronunciation intelligibility

0 1

Similarity CFG Rate

controls similarity to reference audio

0 1

Top-p

AR model sampling top P

0.1 1

Temperature

AR model sampling temperature

0.1 2

Repetition Penalty

AR model sampling repetition penalty

1 3

convert style/emotion/accent

anonymization only

Stream Output Audio

Full Output Audio

Examples

Source Audio	Reference Audio	Diffusion Steps	Length Adjust	Intelligibility CFG Rate	Similarity CFG Rate	Top-p	Temperature	Repetition Penalty	convert style/emotion/accent	anonymization only

Seed Voice Conversion

Seed Voice Conversion V2 (Voice & Style Conversion)

Seed Voice Conversion V1 (Voice & Singing Voice Conversion)