VibeVoice
Free.ai (self-hosted)
·
tts
·
~500 Ii-token clip
·
4.3
I- 3 Abathathi nxaxheba
VibeVoice yi an Umbhalo-usuka-kwilizwi eyenziwe ngu Microsoft. Enamandla kakhulu kwi Long-form audiobooks and multi-speaker podcasts with up to four distinct voices.. I-self-hosted kwi-Free.ai GPUs — iqhuba ngokukhululekileyo kwi-token pool yakho yosuku (500 tokens Iqhosha ngalinye). Ikhutshwe phantsi kwe MIT — ukusetyenziswa korhwebo kuvunyelwe kwi Free.ai.
Sebenzisa nge-API
curl -X POST https://api.free.ai/v1/tts/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"vibevoice","text":"hello world"}'Imibuzo ebuzwa rhoqo
VibeVoice ixhasa uluhlu olubanzi lweelwimi. Uluhlu oluchanekileyo luxhomekeke kwi-engine; ifomu kule phepha ivuma nawuphi na umbhalo kwaye i-engine izakuveza kwiilwimi zayo ezixhaswayo. Bona /voice/ kwi-multi-engine picker epheleleyo ukuba ufuna ulwimi oluthile.
Iinjini ezininzi zibonisa isiNgesi esingenanto- saseMelika ngokumiselweyo kunye nesiqendu esifanelekileyo selizwe leelwimi ezingasazi-siNgesi. Iinjini eziphezulu zingabonisa iziguqulelo zesiqendu - dibanisa isampuli yokuthelekiswa.
Inkxaso ye SSML itshintsha ngokwenjini. Ukuphumla, iprosody, kunye ne tags yokungaqhelekanga zihlonishwa kwinjini ezininzi eziphezulu kunye nakwizinye ezinomphathi- we- self. Umbhalo ocacileyo usoloko usebenza - akukho phawulo lufunekayo.
Ukusasazwa kwe-TTS kufumaneka kwi-premium engines nge /v1/tts/ API endpoint nge stream=true. I-web UI kule phepha ibuyisela i-clip epheleleyo xa ukubonakalisa kugqityiwe.
VibeVoice isebenza kwi-GPU yethu. Uhlobo lutsala kwi-pool yakho yasimahla yansuku zonke kuqala. Xa lugqityiwe, i-token ehlawulweyo iqala kwi-$5 → 200,000 i-token. I-token engaphezulu ~5 nganye yophawu, i-100 encinci nganye ye-clip.
Ukufikelela kwiimpawu ezingama-5,000 ngesicelo ngasinye kwi-web UI. Kwiinxalenye ezide (iincwadi zesandi, izihloko ezipheleleyo), sebenzisa /voice/audiobook/ equka i-chunks ne-stitches ngokuzenzekelayo, okanye thumela i-API kwi-loop.
Ewe — UTHENGA uluhlu lwamagama kwi /v1/tts/batch/, okanye sebenzisa isithuba sokusebenza UI kwi /workspace/ ukudibanisa i-TTS kwindlela ende yokuqhuba (umzekelo, guqulela → thetha → stitch).
Ewe — UTHENGA umbhalo kwi /v1/tts/ ngemodeli "VibeVoice" (okanye i-slug kule phepha). Ibuyisela i-WAV okanye i-MP3. Bona /api/ ukubhekisa okupheleleyo + i-SDK snippets.
Eli phepha liyi-text-to-speech, hayi ukucloning kwelizwi - ilizwi limiselwe kwi-engine. Ukuklona kwelizwi (ukulayisha isandi esibhekisa kuyo), bona /voice/clone/, ekufuneka ube nayo ilungelo lelizwi okanye ube negunya elibhaliweyo elicacileyo.
Iinjini ezimkelweyo ziqhuba kwi-Free.ai- eyenziweyo ye GPUs; akukho nto ishiya iiseva zethu. Iinjini eziphezulu zidlulisa umbhalo kumaqela aphezulu abonelela ngemodeli phantsi kwe DPA yethu. Asiqeqeshi kwiingxelo zakho kwaye asithengi idata.
Ewe — Free.ai inikezela ngenkonzo yorhwebo lwesandi esiveliswe. Ilayisensi ephantsi ye-engine (i-Apache 2.0, MIT, okanye iimeko zomboneleli) ibonisiwe phezulu kunye nephepha lobhekiso lemodeli; kwimisebenzi le kuthetha ukuba i-voiceovers, i-ad, i-podcasts, kunye neenkqubo zekhompyutha zonke zikwi-ambithi.
Ewe - imisebenzi engaphumelelanga ibuyiselwa ngokuzenzekelayo kumbhali (i-pool yosuku okanye i-token ehlawulweyo). Ukuba ubuyiselo alubonakalanga ngaloo mini, thumela i-imeyile kwi contact@free.ai.