Up tókànItọsọna atẹle
Speech Recognition for People with Atypical Speech
Ohùn AI
Audio AI Itọsọna
Whisper is OpenAI’s multilingual speech-recognition model, trained with large-scale weak supervision on approximately 680,000 hours of web-collected audio.
Gbọ si itọsọna yiiAI-ṣe afihan · ~1 min
Its encoder-decoder design supports transcription, English translation, language identification, and timestamps, while accuracy varies by language, recording, and task.
Introduced in 2022, Whisper is a Transformer encoder-decoder trained on approximately 680,000 hours of multilingual audio paired with transcripts or translation text collected from the web. The data is uneven across languages, and performance varies by language and task. Whisper processes audio in 30-second chunks and uses task and language tokens to select transcription, English translation, or timestamped output. Released weights and inference code are under the MIT license.
Ṣiṣẹda awọn iwe afọwọkọ ti o ṣee ṣe wiwa ati awọn akọle fun awọn adarọ-ese ati awọn fidio YouTube
Ṣiṣe awọn ohun elo ipade ifiwe laaye-awọn akọsilẹ ti o ṣe agbejade awọn akopọ lati Sun tabi ohun ohun Ẹgbẹ
Titumọ awọn ifọrọwanilẹnuwo ni ede ajeji taara si ọrọ Gẹẹsi fun awọn oniroyin
Ṣiṣe awọn irinṣẹ iraye si iṣakoso ohun ati ilana fun awọn olumulo ti ko le tẹ
Whisper’s release supports transcription and captioning applications. Real-world performance depends on language coverage, audio quality, accents, domain vocabulary, and evaluation conditions. Measure error rates on representative recordings, with human review where transcripts have consequential use.
Ìwé ìròyìn ọfẹ
Ìtàn AI mẹta tó jẹ́rìí ní gbogbo owurọ́ ọjọ́ iṣẹ́, kọ́wé ní Gẹ̀ẹ́sì tó rọrùn. Ọfẹ́ fún gbogbo àkókò, kò sí ìpolówó.
Ìmeèlì kan ní gbogbo ọjọ́ iṣẹ́. Yọkúrò ní ìkànìyàn ní ìkànìyàn kan. A kò ta tàbí pín àdírẹ́sì rẹ.
Ṣe idanwo fun ara rẹ
Ìdáhùn lẹ́sẹ̀kẹsẹ̀ lórí gbogbo ìdáhùn, àti iwe-ẹ̀rí tó lè pin pẹ̀lú ID tó lè jẹ́rìí lẹ́yìn tí o bá kọ́ ẹ̀kọ́.
Ṣe àtìlẹ́yìn ẹ̀kọ́ AI ọfẹ́. AI Understanding jẹ́ àjọ àìní èrè 501(c)(3) — kò sí ìpolówó, kò sí àkúnya owó, ní gbogbo àkókò. Ṣe ẹbun kan
Whisper is OpenAI’s multilingual speech-recognition model, trained with large-scale weak supervision on approximately 680,000 hours of web-collected audio. Its encoder-decoder design supports transcription, English translation, language identification, and timestamps, while accuracy varies by language, recording, and task.
Whisper ti ni ikẹkọ lori bii awọn wakati 680,000 ti awọn ede lọpọlọpọ, ohun afetigbọ multitask ti a gba lati oju opo wẹẹbu, ipilẹ data ti o tobi pupọ ati oniruuru.
Whisper ṣe iyipada gige ohun iṣẹju-aaya 30 kọọkan sinu ikanni 80 log-Mel spectrogram kan, eyiti awọn ilana fifi koodu Transformer.
Awọn ipo whisper lori awọn ami ami pataki ni ibẹrẹ iyipada ti o sọ fun ede, iṣẹ-ṣiṣe, ati boya lati gbe awọn aami akoko jade.
Whisper jẹ Ayipada koodu-iyipada koodu: oluyipada naa ka spectrogram ati oluyipada ṣe ipilẹṣẹ awọn ami ọrọ ni aifọwọyi.
Ikẹkọ lori awọn oye pupọ ti awọn oriṣiriṣi, ohun afetigbọ gidi-aye (awọn asẹnti, ariwo, ọrọ agbekọja) fun Whisper ni agbara lati inu apoti ti o lagbara laisi atunṣe agbegbe-kọọkan.
Tesiwaju kikọ
Awọn itọsọna diẹ sii ti a yan fun koko yii
Up tókànItọsọna atẹle
Speech Recognition for People with Atypical Speech
Ohùn AI