Amamodeli we-AI

Sifaka amamodeli angcono kakhulu ezwi le-AI elivulekile. Zonke izimodeli zitholakala nge-API ne-web interface.

Amamodeli ombhalo-kuya-kwezwi

Chatterbox
Okushiwo

Uhlu lwezinhlamvu ezingenalutho ezizohlolwa kusuka kusampula encane yokuxhumana, ngezinhlamvu ezingu-23. Imodeli yethu ejwayelekile yokuhlolwa, imisindo e-AI ephikisayo, nemodeli yethu ejwayelekile yokushintshana kwezwi (Chatterbox VC).

Izilimi 23
Uhlu lwezinhlamvu ezibhekiswe 6 kuya ku-60 s
VRAM ~5GB
Ilayisense MIT
Ukuklonywa kwe-Zero-shot Ukushintsha umsindo Izilimi eziningi Ukusetshenziswa kwebhizinisi
Kokoro, Piper, VITS ne Bark

I-Kokoro 82M (Apache 2.0) iyiphutha; i-Piper inikeza kuphela izizwi ezizosiza ukuqeqeshwa kwedatha ukusebenziseka kwezokuhweba; i-VITS ifaka abakhuluma isiNgisi abangu-109 kanye ne-Bark 130 abasetshenzisiwe ezilimi ezingu-13.

Izizwi 316
Izilimi 19
Ama-Engines 4
Layisha phezulu MP3, WAV, OGG, FLAC
Izilimi eziningi Ukusetshenziswa kwebhizinisi kuvunyelwe
Zama umbhalo ukuguqula umsindo
OpenVoice v2

Umbala wethoni oguqula futhi uklonyelisa imodeli. Ingakhethwa futhi ukuguqulwa kwezwi, lapho iChatterbox VC ikhona ngokuzenzakalela ngoba igcina okuningi kwezwi elidingayo.

Uhlu lwezwi lufana nesimo sethu OpenVoice v2: 0.17 to 0.35
Uhlu lwezwi lufana nesimo sethu Chatterbox VC: 0.64 to 0.75
Ilayisense MIT

Uhlobo lomsindo ofanayo wokulingisa okuguqulwe ku-clip ehlukile yomsindo odinga ukulingiswa. Ama-clip amabili ohlelo olufanayo lokulingisa ngempela athola i-score engu-0.76 kuya ku-0.88.

Ukushintsha umsindo Okukhethwa kukho
GitHub

Ukukhuluma-ngokubhaliwe

Whisper Large v3
Okusetshenziswa kakhulu

I OpenAI's open-source speech recognition model. Ibhala amagama angu-100, ngesikhathi sosuku nokuthola amagama ngokuzenzakalela.

Izilimi 100
Amapharamitha 1.5B
VRAM ~10GB
Ukuthola ulwimi Izenzakalelayo
Imodeli ezenzakalelayo Izilimi eziningi Ama-timestamps
Whisper Large v3 Turbo

Incane, iguqulelo elisheshayo le Whisper Large v3 lenza ukudluliswa okusheshayo. Khetha njenge whisper-turbo.

Izilimi 100
Amapharamitha 809M
VRAM ~6GB
Ilayisense MIT
Ekhawulela kakhulu Izilimi eziningi

Imodeli yezingxoxo zomsindo

Qwen2.5 7B Instruct

Imodeli yesilimi esivulekile esibhala iziphenduli kuxoxwa kwezwi, kumuntu okhetha. Isiphequluli sakho sibhala impendulo yakho futhi umsindo we-Piper ufunda impendulo ngokuzwakalayo.

Amapharamitha 7B
Ilayisense Apache 2.0
Umsindo Piper
Isebenza I-GPU yethu
I-Personas ezenzakalelayo
GitHub