Moonshine

STT.ai ma socoto noocan — boggani waa macluumaad ku saabsan ka soo horjeeda. La socodsiiya oo leh moodooyinka aan ku socdaan →
7.8%
WER
1
Afaf
80.0x
Xawaaraha
MIT
Liisan

_Qormo Moonshine

Moonshine waa mid aad u yar oo hadal-to-text ah oo loogu talagalay deegaanka la xaddiday. Marka kaliya 61M parameters, waxay ku socotaa si wax ku ool ah qalabka cidhifka sida Raspberry Pi inta lagu jiro ilaalinta saxnaanta macquulka ah ee Ingiriisiga.

Afaf lagu taageerayo Moonshine

Su'aalaha badanaa la isweydiiyo

Moonshine waa hadal-to-text qaabka by Useful Sensors. STT.ai hadda ma socoto Moonshine — boggani waa macluumaad ku saabsan qaabka. Si loo qoro on STT.ai, qaabka dooranaya wuxuu bixiyaa qoyska Whisper (Turbo, Large V3, Dheer, yar).

On tirakoobyada caadiga ah, Moonshine gaadhay in ka badan 7.8% Word Error Rate. Real-world saxnaanta ku xiran tahay tayada audio, afka, iyo afka; in ay maqal ah ama afka, ka fikiraan boqolkiiba dhibcood yar oo ka sareeya WER.

Moonshine ma aha mid la heli karo STT.ai. Sharciyada liisankiisa ayaa maamula inuu ku socdo adiga. STT.ai's free tier covers the Whisper models we do run - 600 daqiiqo si ay u bilaabaan, ka dibna bishii bilaash ah oo bilaash ah.

Moonshine waxaa lagu soo saaray hoos MIT, a permissive license furan-source. Waxaad awoodi kartaa self-host Moonshine on your hardware ama isticmaali version our martida — labadaba waa ganacsi loo isticmaali karo.

Moonshine taageeraa 1 luqadood. Auto-ka-qabashada doorata luqadda saxda ah ee audio ugu badan; waxaad sidoo kale ku qeexi kartaa gacanta si ay u qaado saxnaanta yar.

Moonshine audio ku saabsan 80.0x waqti dhab ah on our GPUs. A 1-saac file audio dhamaystiran hoos 1 daqiiqo; files dheeri ah oo fariin email ah marka la sameeyo.

Moonshine waxaa ku jira 61M parameters. Models weyn u badan tahay in ay ka sii sax ah laakiin ka sii dhakhso badan; STT.ai martida Moonshine on GPU sidaas darteed tirada parameter ma saameyn ku yeelan doontaa shaqadaaga dhinac macaamiisha.

Moonshine aqbalaa qaab kasta oo STT.ai taageeraya — MP3, WAV, M4A, FLAC, OGG, MP4, MKV, MOV, WebM, AVI, iyo kuwa kale. soo saarka sida TXT, SRT, VTT, DOCX, JSON, ama PDF.

Haa. Speaker diarization la socda Moonshine oo ku saabsan qoraal kasta — hadal kasta oo waxaa lagu calaamadiyay oo waxaad ku bedeli kartaa magacooda editor ka dib.

Haa. Moonshine ku socda deegaankayaga maamula - audio waa la xakamayn karaa oo la tirtiri karaa si default ah oo aan marnaba loo isticmaalin tababarka aan la oggolaanin. Pro qorshayaasha ku darto sirta dhinac macaamiisha ee qoraalada fadhiya.

isticmaali qalab la barbar dhigo-stt si ay u socdaan Moonshine ka hor mid ka mid ah noocyada kale ee taageeray on audio isku mid ah - waxaad arki doontaa WER, qaybta tirada, labels hadal, iyo kalsoonida dhibcaha dhinac-by-laab. The Moonshine vs Whisper Large V3 la barbardhigo waa ugu badan ee caadiga ah la socda.

Haa. Xulo "moonshine" sida paramtirka moodalka ee / v1 / transcribe endpoint. Python iyo Node.js SDKs waxaa ka mid ah Moonshine tusaale. Free API tier waxaa ka mid ah 100 daqiiqo / bilood.

Waa yaabe. Sababtoo ah Moonshine waa MIT-licensed, waxaad awoodi kartaa inaad iska diiwaangeliso. STT.ai's open-source page liiska mashruuca repo iyo miisaanka. Kooxaha wax soo saarka badankood waxay isticmaalaan noocayaga martida ah si ay u dhaafaan GPU iibka, qaababka isbeddelka, iyo ops.
Made by @nadermx