Iimodeli ze-AI

Sidibanisa imodeli elungileyo ye-AI yokuthetha-thethana evulekileyo. Zonke imodeli zifumaneka nge-API kunye nothungelwano lwewebhu.

Iimodeli zombhalo-ukuthetha

Chatterbox
Iinketho zendawo

Uhlobo lwesandi olunombala omnyama omnyama ovela kwisampuli efutshane yobhekiso, kwiilwimi ezili-23. Imodeli yethu emiselweyo yokuluka, ilizwi kwi-AI elibambisayo, nemodeli yethu emiselweyo yokuguqula ilizwi (Chatterbox VC).

Iilwimi 23
Isandi esibhekisa kuyo 6 ukuya 60 imizuzwana
VRAM ~5GB
Ilayisensi MIT
Uklonelo lwe-Zero-shot Uguqulelo lwesandi Iilwimi ezininzi Ukusetyenziswa kwentengiso
I-Kokoro, i-Piper, i-VITS ne-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.

IiNcwadi 316
Iilwimi 19
Iinjini 4
Layisha ezantsi MP3, WAV, OGG, FLAC.
Iilwimi ezininzi Ukusetyenziswa kwentengiso kuvunyelwe
Zama ukuguqula umbhalo ube luvo
OpenVoice v2

Umbala wokuguqulela umbala kunye nemodeli yokukrola. Iyakhethwa ukuguqula ilizwi, apho iChatterbox VC ngoku imiselwe kuba igcina ngaphezu kwelizwi elibekiweyo.

Uthelekiso lwelizwi kuvavanyo lwethu OpenVoice v2: 0.17 to 0.35
Uthelekiso lwelizwi kuvavanyo lwethu Chatterbox VC: 0.64 to 0.75
Ilayisensi MIT

Uthelekiso lomculi wokulinganisa okuguqulweyo kokulinganisa okuguqulweyo kumfanekiso ongumqondiso ohlukileyo womculi omiselweyo. Iividiyo ezimbini zomculi ofanayo wenyaniso zifumana inqaku le-0.76 ukuya kwi-0.88.

Uguqulelo lwesandi Ekunokukhethwa kuko
GitHub

Iimodeli zokuthetha- ukuya- kumbhalo

Whisper Large v3
Esetyenziswa kakhulu

I-OpenAI's open-source speech recognition model. Iguqula iilwimi ezili-100, ngee-timestamps nobuchwephesha obuzenzekelayo bokukhangela iilwimi.

Iilwimi 100
Iiparamitha 1.5B
VRAM ~10GB
Ubhaqo lwesiNgesi Iinketho ze projekti
Imodeli Emiselweyo Iilwimi ezininzi Ii-Timestamps
Whisper Large v3 Turbo

Uhlobo oluncinci, olukhawulezayo lwe Whisper Large v3 olunokuguqulelwa ngokukhawuleza. Khetha njenge whisper- turbo.

Iilwimi 100
Iiparamitha 809M
VRAM ~6GB
Ilayisensi MIT
Ekhawulezayo Iilwimi ezininzi

Imodeli yencoko yelizwi

Qwen2.5 7B Instruct

Imodeli yoLwimi oluvulekileyo olubhaliweyo kwincoko yelizwi, kwipersona okhethayo. Umkhangeli zincwadi wakho uguqula uthetha kwakho kwaye ulizwi le Piper lifunda impendulo ngokunzulu.

Iiparamitha 7B
Ilayisensi Apache 2.0
Ilizwi Piper
Isebenza I-GPU yethu
Iinkcukacha
GitHub