VibeVoice
Free.ai (self-hosted)
·
tts
·
~500 Ama-token ngalinye clip
·
4.3
kusuka 3 abasebenzisi balolu hlu
VibeVoice yi an umbhalo-kuya-kwezwi eyenziwe ngu Microsoft. Onamandla kakhulu ku Long-form audiobooks and multi-speaker podcasts with up to four distinct voices.. I-self-hosted ku-Free.ai GPUs - isebenza ngokukhululekile ngaphezu kwe-token pool yakho yansuku zonke (500 tokens i-clip ngayinye). Kukhishwa ngaphansi kwe-MIT — ukusetshenziswa kokuthengiswa kwempahla kuvunyelwe ku-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 kaningi
VibeVoice isekela izilimi eziningi. Uhlu oluphelele luya ngonjini; ifomu kulekhasi livuma noma yiluphi umbhalo futhi injini izoveza ngezilimi zayo ezixhasiwe. Bona /voice/ ukukhetha okuphelele kwenjini eminingi uma ufuna ulwimi oluthile.
Injini eziningi zinikeza isiNgisi se-American-neutral ngokuzenzakalela kanye nendawo efanele yendawo ye-English. Injini eziphezulu zingaveza izinketho zesiNgisi - chofoza isibonelo ukulinganisa.
Insizakalo ye-SSML ihluka ngokwenjini. Ukuyeka, i-prosody, kanye ne-emphasis tags zihlonishwa kuningi kwama-premium engines kanye namanye ama-hosted self. Umbhalo ocacile usebenza njalo — akukho uphawulo oludingekayo.
Ukusakazwa kwe-TTS kutholakala kuma-engine aphezulu nge /v1/tts/API endpoint nge-stream=true. I-web UI kule khasi ibuyisela i-clip ephelele uma ukusakazwa kuqediwe.
VibeVoice isebenza kuma-GPUs ethu. Ukuzalwa kuqala kuvela epulazini lakho elimahhala ngosuku lokuqala. Uma liphelelwe, ama-token akhokhelwayo aqala ku-$5 → ama-token angama-200,000. Kubukeka sengathi ~5 ama-token ngayinye yophawu, okuncane 100 ngayinye ye-clip.
Kuze ku-5,000 amaphawu ngesicelo ngasinye kwi-web UI. Ukusebenzisa amathuluzi angaphezu kwesinye (ama-audiobooks, amasigaba agcwele), sebenzisa /voice/audiobook/ okungukuthi, ama-chunks nama-stitches ngokuzenzakalela, noma thinta i-API nge-loop.
Yebo — POST uhlu lwama-strings ku /v1/tts/batch/, noma sebenzisa indawo yokusebenzela UI ku /workspace/ ukudlulisa i-TTS ku-pipeline ende (isibonelo, guqula → khuluma → stitch).
Yebo — IKHAYA lombhalo ku /v1/tts/ ngemodeli="VibeVoice" (noma i-slug kulekhasi). Ibuyisela i-WAV noma i-MP3. Bona /api/ ukubhekisa okugcwele + ama-SDK snippets.
Lekhasi liyi-text-to-speech, hhayi ukuklonya umsindo — umsindo yiphutha lenjini. Ukuklonya umsindo (ukufaka umsindo obhekiswe kuwo), bona /voice/clone/, okudinga ukuthi ube nelungelo lokukhuluma noma ube negunya elibhalwe phansi.
Injini ezihlala ngokwazo zisebenza ku-Free.ai-owned GPUs; akukho okushiya amaseva ethu. Injini eziphezulu zidlulisa umbhalo kumamodeli aphezulu ahlinzekwa ngezansi kwe-DPA yethu. Asiqeqeshi ngemingeniso yakho futhi asithengi idatha.
Yebo — Free.ai inikeza ukusetshenziswa komphakathi kwesandi esikhiqizwe. Ilayisense elingaphansi kwenjini (i-Apache 2.0, MIT, noma imithetho yomthengisi) liboniswe phezulu kanye nekhasi lo referensi lemodeli; ngokusebenza lokhu kusho ukuthi izizwi, izikhangiso, amapodcast, nama-apps wonke atholakala ngaphakathi kwe-scope.
Yebo — imisebenzi engaphumelelanga ibuyiselwa ngokuzenzakalela emthonjeni (i-pool yansuku zonke noma ama-token akhokhelwe). Uma ubuyiselwa kwemali akubonisi ubusuku obufanayo, thumela i-imeyili contact@free.ai.