Moonshine Base
Free.ai (self-hosted)
·
stt
·
~500 Ama-token ngalinye minute
Moonshine Base yi an imodeli yokukhuluma-ku-umbhalo eyenziwe ngu Useful Sensors. Onamandla kakhulu ku Low-latency live transcription, embedded devices.. I-self-hosted ku-Free.ai GPUs - isebenza ngokukhululekile ngaphezu kwe-token pool yakho yansuku zonke (500 tokens imizuzu). Kukhishwa ngaphansi kwe-MIT — ukusetshenziswa kokuthengiswa kwempahla kuvunyelwe ku-Free.ai.
Sebenzisa nge-API
I-REST API ehambisana ne-OpenAI. Dala isithonjana bese ubiza le modeli emaminithini.
curl -X POST https://api.free.ai/v1/stt/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"moonshine-base","audio_url":"https://..."}'Imibuzo ebuzwa kaningi
Moonshine Base iguqula umsindo okhuluma ube ngumbhalo. Layisha phezulu ihele le-MP3, WAV, M4A, noma levidiyo futhi Moonshine Base ibuyisela umbhalo ophelele kanye ne-SRT/VTT ekhethiwe enezihloko ezingezansi ezinezikhumbuzo zesikhathi.
Moonshine Base iphatha izilimi eziningi — Whisper-imindeni yemodeli ifaka 90+, Parakeet ifaka ~25, abanye bahluka. Khetha "ukuthola ngokuzenzakalela" noma chaza ulwimi ukuze kube lula kakhulu.
Izinga lephutha legama lingu-5-10% ku-English audio ehlanzekile, 10-20% ku-noisey noma accented audio. Ama-variants amakhulu wesakhiwo esifanayo enza kahle kakhulu kulezi zinkinga ezinzima — khetha okukhulu uma i-audio inzima.
Yebo — yonke inqwaba ifaka iqala/iphele ama-timestamps. Rhweba ngaphandle njenge-SRT noma i-VTT futhi ama-timesmap ahamba ngqo kwividiyo yakho.
Moonshine Base isebenza kuma-GPU ethu ahlukile eqenjini lakho elimahhala ngosuku lokuqala; $5 → 200,000 ama-token akhokhelwe ngemuva kwalokho. Ama-token angama-500 ngomzuzu.
MP3, WAV, M4A, FLAC, OGG, kanye nevidiyo (MP4, MOV, WebM) — sikhipha umsindo. Max 500 MB ngayinye. Amafayela ade kakhulu? Yahlula nge /audio/cut/ noma sebenzisa /v1/stt/batch/.
Ukudweba umsindo womsindo kudlula ngokuhlukile - shintsha "ukudweba" ku /transcribe /. Moonshine Base iphatha ukudweba; ukudweba kufaka izixhumanisi kunoma iyiphi inhlanganisela yomsindo womsindo 1 / umsindo 2 / njll.
Yebo — /batch/ i accetta ifolda yefayela le-audio. I-transcript ngayinye iwela ku /account/?tab=history ngegama lefayela elisemthethweni. Ukugcina ifolda-isihlahla sebenzisa i-API.
Yebo — UTHENGA umsindo wakho ku /v1/stt/transcribe/ ngemodeli "Moonshine Base". Ibuyisela i-JSON ngemibhalo + amasegmenti + ama-timestamps egama-level. /api/ ine-reference ephelele.
Amamodeli ahlala ahlala agcina umsindo ku-GPUs ethu; ama-premium adlula nge-DPA. Umsindo ususwa ngemuva kohlelo lokuhlukaniswa kwefasitela (24h anon, 7d ukungena). Asiqeqeshi ngemingeniso yakho.
Yebo — Free.ai inikeza ukusetshenziswa kokuthengiswa kwezincwadi. Udinga izivumelwano zesandi esilayishiwe (ukufaka kwakho, izinto ezilayisense, noma okuqukethwe ngesivumelwano).
Isikhathi sangempela siyi-0.05-0.2× — i-podcast yemizuzu engu-60 ibhala imizuzwana engu-3-12. Amamodeli aphezulu avame ukuqedela ngokushesha. Sebenzisa inkinobho yokulinganisa ukuvala i-tab.