We use 3-7 seconds (or five utterances) of the target speaker to clone their voice, then generate fully fake utterances. Partially fake utterances were created by combining the bona fide and the fully fake.
*Loudness were normalized for demonstration purposes
aLJ JETS is not a voice-cloning system but a single-speaker TTS model with the voice of a female speaker.
SOURCE:
•VILLEFORT•CONTINUED•HE•KNOWS•ME••AND•I•••HAVE•PLEDGED•MY•WORD•TO•HIM•
TARGET:
•VILLEFORT•CONTINUED•HE•KNOWS•HER•AND•SHE•HAS••PLEDGED•MY•WORD•TO•HIM•
| Female (84) | Fully Fake | Partially Fake | ||
|---|---|---|---|---|
| Crossfade | Cut/Paste | Overlap/Add | ||
| Bona fide | â–ş Play 84_121123_000050_000005.wav | |||
| LJ JETSa | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| YourTTS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| XTTS v2 | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| GPT-SoVITS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| Cosy Voice | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| ElevenLab | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
SOURCE:
•THE•POINT•OF•VIEW•IN•WHICH•THIS•TALE•COMES•UNDER•THE•ROMANTIC•DEFINITION•LIES•IN•THE•ATTEMPT•TO•CONNECT••••A•BYGONE•TIME•WITH•THE•VERY•••••PRESENT•THAT•IS•FLITTING•AWAY•••••FROM•US•
TARGET:
•THE•POINT•OF•VIEW•IN•WHICH•THIS•TALE•COMES•UNDER•THE•REALIST••DEFINITION•LIES•IN•THE•ATTEMPT•TO•DISCONNECT•A•BYGONE•TIME•FROM•THE•STAGNANT•PAST••••THAT•IS••••••••••STICKING•WITH•US•
| Male (2086) | Fully Fake | Partially Fake | ||
|---|---|---|---|---|
| Crossfade | Cut/Paste | Overlap/Add | ||
| Bona fide | â–ş Play 2086_149214_000005_000001.wav | |||
| LJ JETSa | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| YourTTS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| XTTS v2 | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| GPT-SoVITS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| Cosy Voice | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| ElevenLab | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
SOURCE:
•THE•WORSHIP•••OF•THE•GREAT•••MYSTERY•WAS•SILENT•••SOLITARY•FREE•••FROM•ALL•SELF•••••SEEKING•
TARGET:
•THE•REVERENCE•OF•THE•ANCIENT•ENIGMA••WAS•TRANQUIL•ISOLATED•PURGED•OF•••ALL•EGOISTIC•DESIRES•
| Male (5536) | Fully Fake | Partially Fake | ||
|---|---|---|---|---|
| Crossfade | Cut/Paste | Overlap/Add | ||
| Bona fide | â–ş Play 5536_43358_000011_000000.wav | |||
| LJ JETSa | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| YourTTS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| XTTS v2 | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| GPT-SoVITS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| Cosy Voice | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| ElevenLab | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
SOURCE:
•AS•SOON•AS•IT•WAS•ENDED•THEY•PROCEEDED•TO•OVERHAUL•MY•SWAG•AND•THE•CONTENTS•OF•MY•POCKETS•
TARGET:
•AS•SOON•AS•IT•••••BEGAN•THEY•FAILED••••TO•TOUCH••••MY•SWAG•OR••THE•CONTENTS•OF•MY•POCKETS•
| Female (2412) | Fully Fake | Partially Fake | ||
|---|---|---|---|---|
| Crossfade | Cut/Paste | Overlap/Add | ||
| Bona fide | â–ş Play 2412_153954_000008_000009.wav | |||
| LJ JETSa | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| YourTTS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| XTTS v2 | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| GPT-SoVITS | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| Cosy Voice | â–ş Play | â–ş Play | â–ş Play | â–ş Play |
| ElevenLab | â–ş Play | â–ş Play | â–ş Play | â–ş Play |