Seed-VC
Seed-VC is named after Seed-TTS as the method and training scheme is inspired by the paper's description about voice conversion.
We may change name based on future development of this project.
This page is for showing results and comparisons only. The examples of other VC models are directly retrieved from their demo pages.
Zero-shot Voice Conversion
We present our results of zero-shot voice conversion comparing to other previous released/published voice conversion models.
We demonstrate our model has comparable if not superior performance to the previous models. The results are shown below.
| Source Audio | Timbre Prompt | Mulli VC | Consistency VC | DiffHier VC | Free VC | Seed-VC (ours) |
|---|---|---|---|---|---|---|
Celebrities and Anime Characters
Seed-VC can mimic the voices of celebrities or characters from animated shows. We present these examples for purely research purposes.
| Source Audio | Celeb | Timbre Prompt | Seed-VC Converted |
|---|---|---|---|
| Donald Trump | |||
| Benedict Cumberbatch | |||
| 周杰伦 (Jay Chou) | |||
| 丁真 (Zhen Ding) |
| Source Audio | Anime/Game Character | Timbre Prompt | Seed-VC Converted |
|---|---|---|---|
| Rick (from Rick and Morty) | |||
| Morty (from Rick and Morty) | |||
| 后羿 (Houyi) | |||
| 八重神子 (Yae Miko) | |||
| Glados (from Portal) |
Talking Pace Control
Seed-VC can easily control the talking pace of the converted audio.
| Source Audio | Timbre Prompt | Length | Seed-VC Converted |
|---|---|---|---|
| x0.5 | |||
| x1.0 | |||
| x1.5 | |||
| x2.0 |
Singing Voice Conversion
Seed-VC perform singing voice conversion as well. The results are shown below.
| Timbre Prompt | Source Audio | Seed-VC Converted |
|---|---|---|
VS Seed-TTS Demo
We compare our result with the ones presented in Seed-TTS's demo page.
The current released checkpoint has reached close to the performance of the Seed-TTS's VC model. The results are shown below.
| Language | Source Audio | Timbre Prompt | Seed-TTS | Seed-VC (ours) |
|---|---|---|---|---|
| EN | ||||
| ZH | ||||