Yiddish e mahala

Yiddish ke puo e nang le li-resource tse tlase bakeng sa Whisper - large-v3-turbo e lula ka holimo ho 25% ea palo ea liphoso tsa mantsoe, ha ho tluoa ho feta. Li-transcript li ka sebelisoa ho batla le ho ngola empa ha li lokela ho nkoa e le li-publishing-ready. Haeba e-enjine e phahameng ea ho nepahala e ka fumaneha bakeng sa Yiddish re e hokahanya ka ho toba.(Tier D, over 25% word error rate ka lihlopha tsa benchmark - re phatlalatsa li-tiers tsa WER tse tšepahalang ho feta li-claims tsa thekiso.)

E-na — Yiddish transcription e nka ho tloha ho token pool ea hau ea mahala ea letsatsi le letsatsi pele. Audio e theko e tlaase ka li-token tse ka bang 50 ka metsotsoana, ka hona pool ea letsatsi le letsatsi e sa tsejoeng e koahela lihora tse'maloa tsa audio ka letsatsi. Li-account tse kentsoeng li fumana pool e kholo ea 30,000-token ka letsatsi. Ka mor'a moo, transcription e lefa-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-

Li-transcripts tsa Yiddish li khutlisoa ka UTF-8 e tloaelehileng le ho ngola ka mokhoa o tloaelehileng oa puo.

MP3, WAV, M4A, FLAC, OGG, OPUS, le WEBM li amoheloa ka kotloloho. Bakeng sa video (MP4, MOV, MKV) re tlosa li-track tsa audio ka lehlakoreng la mosebelisi pele re li romella ho Whisper — ha o hloka ho fetola ntho efe kapa efe ka boeona. Pipeline e ts'oanang ntle le puo ea mohloli, ho kenyeletsoa Yiddish.

Li-account tse kentsoeng ka har'a li-account li fihla ho 2 GB. Nako e telele ha e lekanyetsoe - li-file tse telele li kopanngoa ka ho toba (lifensetere tsa metsotsoana e 30 le ho koaloa)'me li kopanngoa ho ea ho transcript e le' ngoe le li-timestamps tse tsoelang pele. Li-recording tsa Yiddish tse fetang lihora tse ngata (podcasts, lihlooho tse felletseng, liboka) li sebetsa hantle.

Ha ho joalo — ho arolelana lingoliloeng tsa moqoqo ho hokahantsoe ka ho sa feleng bakeng sa Yiddish transcript e ngoe le e ngoe. Liphetho li arotsoe e le Moqoqo 1 / Moqoqo 2 / Moqoqo 3 le li-timestamps, ka hona lipotso tsa puisano, li-panel discussions, le liketsahalo tse ngata li khutlela ka letšoao. Ho arolelana lingoliloeng tsa moqoqo ho sebetsa ka mokhoa o fapaneng'me ho sebetsa ka tsela e ts'oanang ka lipuo tsohle tseo re li tšehetsang.

E-ea — kopanya URL ho /transcribe/youtube/ bakeng sa YouTube kapa /transcribe/podcast/ bakeng sa podcast feeds (Apple, Spotify, RSS). Re kenya audio, e tsamaisa ka Whisper le language=yi,'me re khutlisetse transcript le timestamps le li-labels tsa moqolotsi oa litaba. Lintlha tsa Yiddish: Lingoliloeng, lipotso, litlhaloso tsa puisano, le litaba tsa YouTube ka Yiddish li sebetsa — kenya URL ho /transcribe/youtube/ kapa u romelle faele ka kotloloho.

Whisper e theko e tlaase ka li-token tsa 50 ka metsotsoana ea audio, ka hona ho ngolisoa ha hora e le 'ngoe ke li-token tsa ~ 3,000. Basebelisi ba bangata ha ba sebelise ntho e ngoe le e ngoe - pool ea letsatsi le letsatsi ea li-token tsa 30,000 e kenyelletsa li-clip tse khuts'oane, li-notes tsa molumo le podcasts tse le' ngoe. Kantle ho eona, ho ngolisoa ke ho lefa-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-na-ka-ho-e-

Ee — li-timestamps tsa lefatše la sehlooho (ka ~10-30 sekontiri) le tsa lefatše la mantsoe li fumaneha. Lefatše la mantsoe ke la morao-rao bakeng sa ho romelloa ha li-subtitles tsa VTT/SRT ka hona li-captions li hokahana line-by-line. Ka API beha timestamps="word" ka'mele oa kopo. Li-transcripts tsa Yiddish li khutlisoa ka UTF-8 e tloaelehileng le ho ngola ka mokhoa o tloaelehileng oa puo.

Ee. POST audio (multipart/form-data, field name "file") ho /v1/transcribe/ le language=yi — kapa u se ke ua beha parameter ea puo ho etsa hore Whisper e fumane ka ho toba. E khutlisa JSON le transcript, segments, timestamps, le labels tsa motsamaisi. Litlhahlobo tse felletseng le li-snippets tsa SDK li ka fumaneha ho /api/.

E-ea — ha transcription e felile, tobetsa ho fetolela kapa ho kopanya tekanyo ho /translate/. Yiddish e kopantsoe le lipuo tsohle tse ling tseo re li tšepang (200+). Bakeng sa lihora tsa kopano, transcript e fetisoa ka /summarize/; bakeng sa ho ngola ka letsoho, e romelle ho /voice/tts/ ho etsa molumo ka puo e loketseng.

Ho ithuta ho utloa mollo oa Whisper ho thusa ka tlase ho lenane lena — bothata ke palo ea Yiddish audio Whisper e e bonang ha e ithuta, eseng mollo. Mollo o hloekileng oa studio o ntse o hlola mollo o moholo, empa o ke ke oa fihla ho hlaka eo u ka e fumanang ka puo e nang le li-resource tse ngata.Haeba transcript e sa khone ho sebelisoa, romella lengolo-tsoibila contact@free.ai le faele — re tla lefa tokens'me re sheba hore na e mong oa li-engine tse fapaneng o sebetsana joang le audio ea hau ka ho fetisisa.