Li-Models tsa AI

Re hokahanya li-models tse ntlehali tsa ho bua ka AI. Li-models tsohle li fumaneha ka API le interface ea web.

Li-Models tsa Tebe-ho-Bongata

Chatterbox
E khothalletsoang

Ho hloa ha lentsoe la Zero-shot ho tloha ho mohlala o khuts'oane oa ho ngola, ka lipuo tse 23. Mofuta oa rona oa ho hloa ka ho fetesisa, li-voices ka AI doubling, le mofuta oa rona oa ho fetola lentsoe ka ho fetesisa (Chatterbox VC).

Li-languages 23
Li-audio tsa ho ngola 6 ho isa ho 60 sec
VRAM ~5GB
Laesense MIT
Cloning Zero-shot Ho fetola puo Li-languages tse ngata Ho sebelisa ka khoebo
Kokoro, Piper, VITS le Bark

Four open text to speech engines behind one voice picker. Kokoro 82M (Apache 2.0) is the default; Piper offers only voices whose training data allows commercial use; VITS adds 109 English speakers and Bark 130 presets in 13 languages.

Li-voices 316
Li-languages 19
Li-Engines 4
Kopitsa MP3, WAV, OGG, FLAC
Li-languages tse ngata Ho sebelisoa ka khoebo ho lumelloa
Tlanya ho fetola mongolo ho ho bua
OpenVoice v2

Mofuta oa ho fetolela le ho klona. Ho ntse ho ka khethoa bakeng sa ho fetolela ha molumo, moo Chatterbox VC e leng ka ho sa feleng ka hona e boloka molumo o loketseng haholo.

Ho lumellana ha lentsoe ho tlaleha ho rona OpenVoice v2: 0.17 to 0.35
Ho lumellana ha lentsoe ho tlaleha ho rona Chatterbox VC: 0.64 to 0.75
Laesense MIT

Ho tšoana ha moqoqo oa ho ngolisoa ho fetotsoeng ho clip e fapaneng ea moqoqo oa sepheo. Li-clip tse peli tsa moqoqo o tšoanang oa sepheo li na le sekhahla sa 0.76 ho isa ho 0.88.

Ho fetola puo E ka ba teng
GitHub

Li-Models tsa ho bua ho ea ho tokiso

Whisper Large v3
E sebelisoang haholo

Mofuta oa OpenAI oa ho tseba mantsoe a nang le lifaele tse bulehileng. E ngola lipuo tse 100, ka li-timestamp le ho bona lipuo ka ho toba.

Li-languages 100
Liparametara 1.5B
VRAM ~10GB
Ho fumana puo Ka boiketsetso
Mofuta o tloaelehileng Li-languages tse ngata Li-timestamps
Whisper Large v3 Turbo

E le mofuta o nyane, o potlakileng oa Whisper Large v3 bakeng sa li-transcripts tse potlakileng. Khetha e le whisper-turbo.

Li-languages 100
Liparametara 809M
VRAM ~6GB
Laesense MIT
Ho potlaka Li-languages tse ngata

Mofuta oa puisano ea lentsoe

Qwen2.5 7B Instruct

Mofuta oa puo ea open-source e ngolang likarabo ho tšebelisano ea molumo, ka motho eo u mo khethang. Moenjiniere oa hau o ngola puo ea hau'me molumo oa Piper o bala karabo ka bongata.

Liparametara 7B
Laesense Apache 2.0
Moqoqo Piper
E sebetsa GPUs tsa rona
Li-Personas tsa hau
GitHub