Kani TTS 2

Kani TTS 2 TTS

An ultra-lightweight 400M English model that runs in just 3GB of VRAM at a 0.2 real-time factor.

Bhalisa Uluhlu lwezinto zobumnini Zolwaleko...

Ulawulo oluchanekileyo:

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

Ii-tags imodeli ekhethiweyo iqonda - nqakraza ukushiya enye kumbhalo wakho apho isenza khona:

Le modeli ifunda umbhalo oqhelekileyo, ngoko ke i-inline tags ilahleka. Uphawu olusekelwe kwi-emotions, tshintshela kwimodeli ebonisa umbono njenge-Orpheus okanye i-Bark.

Chaza ubeko lwephepha

-12 +12
0.5x 2.0x
Ikhululekile nge Piper, VITS, MeloTTS
Isandi sakho esivelisweyo siza kuvela apha. Khetha imodeli, ngenisa umbhalo, kwaye unqakraze Yenza.
Isandi Sizaliswe Ngempumelelo
0:00
Layisha ezantsi Layisha ezantsi Ikhonkco liphelelwe lixesha kwiyure ezi-24
Inqanaba elikhululekileyo: ukusetyenziswa komuntu siqu. Ilayisensi yezorhwebo ukusuka kwi- $5/inyanga
Uthando TTS.ai? Nceda utshele abalandeli bakho!

I-About Kani TTS 2

Kani-TTS-2 by NineNineSix is an ultra-lightweight 400M-parameter text-to-speech model built on a Liquid AI LFM2 backbone with NVIDIA's NanoCodec. It runs in just 3GB of VRAM and generates roughly ten seconds of speech in about two seconds on an A100 — a real-time factor near 0.2. The current public release ships an English-only checkpoint and, unlike its predecessor, does not expose the speaker-embedding hook needed for voice cloning. Its strength is fast, low-cost English generation on modest hardware, which makes it a good fit for quick previews and high-volume English narration. It is released under Apache 2.0 and offered on the free tier.

Elungileyo: Fast English generation on low-VRAM hardware, quick previews

Khangela konke Kani TTS 2 iilizwi

Kwingxelo

Umbhekisi phambili
NineNineSix
Ilayisensi
Apache 2.0
I-Tier
free
Isantya
fast
Ukuphinda usebenzise ilizwi
Akukho nanye
Iilwimi
English
Ubukhulu bamagama
1000

Kani TTS 2 iilizwi

Default

English
Emiselweyo Neutral

Kani TTS 2 TTS - Imibuzo ebuzwa rhoqo

It runs in just 3GB of VRAM and produces about ten seconds of speech in roughly two seconds on an A100 — a real-time factor near 0.2 — thanks to its 400M-parameter LFM2 backbone and NanoCodec.

No. The current v2 release removed the public speaker-embedding hook, so cloning is not available. For cloning, use Chatterbox, IndexTTS-2, or GPT-SoVITS.

English only. The public release ships a single English checkpoint; for non-English speech, use a model like Kokoro or MeloTTS.
← Zonke iingoma