Umwandiko Kuri Abagenzuzi Indirimbo Kuri Umwandiko Guca umurongo Igishushanyo mbonera Ibiro Birenzeho
/
Darwin TTS

Darwin TTS Ikiganiro

A Qwen3-TTS variant whose talker FFN weights are blended from the Qwen3 language model for sharper cross-lingual cloning.

0/500 Inyuguti · Hejuru ya: →
Kwiyandikisha kugirango Inyuguti

Umwandiko in Itagi: ya: Igenzura:

<speak><prosody rate="slow">Slow speech</prosody></speak>

i Byahiswemo Urugero - Kanda Kuri Gukuraho Rimwe Umwandiko:

Urugero: Bisanzwe Umwandiko, Umurongo: Itagi:. Itagi: -, Hindura Kuri Urugero: Nka Cyangwa.

Kugena (Ijambo =):

-12 +12
0.5x 2.0x
Na:,,
Audio Kugaragara. A Urugero:, Injiza Umwandiko, na Kanda.
Byaremwe
0:00
Iyimura Iyimura Ihuza in
Bwite Koresha Kuva: 5 /
TTS.ai? Abayobozi!

Ikiganiro Darwin TTS

Darwin-TTS-1.7B-Cross by FINAL-Bench is a research variant of Qwen3-TTS-1.7B with an unusual construction: 84 of its talker-FFN tensors (about 8.6% of them) are blended at a 3% ratio with the matching tensors from Qwen3-1.7B-Base, all without any retraining. The result is a model that produces noticeably crisper cross-lingual voice cloning across Korean, English, Japanese, and Chinese — its four core languages. It operates in zero-shot voice-clone mode, needing only about three seconds of reference audio to capture a speaker. Darwin is best suited to transferring a single reference voice across those four languages, for example dubbing or multilingual narration with consistent speaker identity.

kugirango: Cross-lingual voice cloning between English / Korean / Japanese / Chinese with a single reference voice

Gushakisha byose Darwin TTS Amashusho

A

Mukoraporogaramu
FINAL-Bench
Inyandiko y'Iyemererakoresha
Apache 2.0
Itariki
standard
Umuvuduko
medium
Guhindura izina
Oya
Ururimi:
English, Korean, Japanese, Chinese
Inyuguti
2000

Darwin TTS Amashusho

Default

English
Bisanzwe Neutral

Default (Chinese)

Chinese
Bisanzwe Neutral

Default (Japanese)

Japanese
Bisanzwe Neutral

Default (Korean)

Korean
Bisanzwe Neutral

Darwin TTS -

Darwin starts from Qwen3-TTS-1.7B but blends a small fraction of its talker-FFN weights with the matching weights from the Qwen3-1.7B base language model. This training-free blend sharpens cross-lingual voice cloning rather than changing the base voices.

English, Korean, Japanese, and Chinese. The FINAL-Bench release specifically markets its cross-lingual blend for those four, and the deployed model ships voices for them.

About three seconds. It works in zero-shot mode, so no fine-tuning or training is required — you provide a short reference clip and it generates new speech in that voice.
← Amashusho yose