Ubwubatsi bwimbitse
DeepSpeech nicyitegererezo cyo kumenyekanisha imvugo iherezo kugeza ku ndunduro yatangijwe na Baidu muri 2014 ishushanya ikarita yerekana amajwi mbisi ku buryo bwanditse ukoresheje umuyoboro w’imitsi usubiramo watojwe nigihombo cya CTC.
Incamake
It helped pioneer the shift away from complex, hand-engineered ASR pipelines toward learned, data-driven systems.
Kwibira cyane
Kumenyekanisha imvugo ya kera byashushanyije hamwe moderi zitandukanye za acoustique, inkoranyamagambo, hamwe nururimi rwindimi hamwe nibice byahujwe n'intoki. DeepSpeech yasimbuye ibyinshi muribyo numuyoboro umwe wimyitozo yatojwe kurangiza. Ubwubatsi bwayo bufata spekrogramu cyangwa MFCC ibiranga hejuru yama majwi magufi kandi ikabigaburira binyuze mubice byinshi byahujwe byuzuye, ibyerekezo byisubiramo byombi bifata imiterere kuva kera ndetse nigihe kizaza, hamwe nibisohoka bitanga amahirwe yo gukwirakwiza inyuguti kuri buri ntambwe. Byibanze, ikoresha Connectionist Temporal Classification (CTC), ituma umuyoboro wiga guhuza amajwi n'inyandiko udakeneye ibirango-urwego. Nyuma Mozilla yasohoye ibyamamare bifungura-isoko (hamwe na verisiyo nshya ukoresheje LSTM ishingiye, igishushanyo mbonera), bituma inzira igerwaho cyane.
Ubushishozi
Urufunguzo rufasha ni igihombo cya CTC. Imvugo ninyandiko ntabwo bihujwe kumurongo-kumurongo, CTC rero itangiza ikimenyetso 'cyambaye ubusa' hamwe nigiteranyo hejuru yuburyo bwose bushoboka bushobora gusenyuka kubigenewe. Ibi bireka icyitegererezo gisohora inyuguti mugihe cyintambwe hanyuma ukamenya aho amajwi ikarita yinyuguti byikora. Ibyerekezo bibiri RNN itanga buri guhanura kugera kumurongo wa acoustic, kandi imvugo n-garama yo hanze ikunze kongerwaho mugihe cya decode kugirango tunonosore imyandikire nijambo.
Ingaruka z'Ingamba
Kugera no kugera
Itezimbere kugerwaho binyuze mu kwandukura, kuvuga, no guhuza amajwi.
Igiciro na bije
Amatsinda yibitangazamakuru arashobora kohereza amajwi yihuse hamwe na bije nto.
Umuvuduko n'igipimo
Sisitemu ireba abakiriya irashobora gutunganya imikoranire ivugwa murwego runini.
Ejo hazaza h'ubwubatsi bwimbitse
DeepSpeech ubwayo yasimbuwe ahanini nubwitonzi- hamwe na transformateur ishingiye ku bwubatsi (Conformer, Whisper, wav2vec 2.0) ifata imiterere ndende no kwiyobora ku majwi atanditse. Ariko ibitekerezo byayo byingenzi, amahugurwa arangira-arangije na decode ya CTC, akomeza kuba urufatiro kandi aracyagaragara muri sisitemu ya kijyambere. Umurage ni igitekerezo: byerekanaga ko icyitegererezo kimwe cyize gishobora guhangana n'imiyoboro ikozwe cyane, bigatanga inzira kuri moderi nini, indimi nyinshi, yiyobora wenyine.
Gushyira mu bikorwa Isi
Offline, kumurongo wibikoresho byamajwi kumenyekanisha porogaramu yibanga ukoresheje Mozilla ifunguye DeepSpeech
Gukora inyandiko-mvugo ya podcasts cyangwa ibiganiro udashingiye kuri serivise
Kwigisha ishingiro ryigihombo cya ASR na CTC mu masomo yo kwiga imashini za kaminuza
Kubaka amajwi yihariye ya IoT cyangwa ibikoresho byashizwemo aho bikenewe byoroshye, byoroshye kumenyekana
Ingaruka & Kurinda
Gukoresha nabi amajwi no kwigira ibyago byiyongera mugihe uruhushya rubuze.
Ukuri kurashobora kugabanuka hejuru yimvugo, imvugo, cyangwa urusaku rwibidukikije.
Amajwi yubukorikori arashobora kwibeshya kumvugo yukuri nta kirango gisobanutse.
Igishushanyo mbonera
Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.
Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.
Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.
Andika amajwi yubukorikori kandi ugumane inyandiko zerekana kubazwa.
Komeza Ubushakashatsi
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the DeepSpeech Architecture quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ubuyobozi bukurikira
Ubwubatsi
Ibibazo bikunze kubazwa
What is DeepSpeech Architecture?
DeepSpeech nicyitegererezo cyo kumenyekanisha imvugo iherezo kugeza ku ndunduro yatangijwe na Baidu muri 2014 ishushanya ikarita yerekana amajwi mbisi ku buryo bwanditse ukoresheje umuyoboro w’imitsi usubiramo watojwe nigihombo cya CTC. Yafashaga abapayiniya kuva mu miyoboro igoye, yakozwe n'intoki ASR yerekeza kuri sisitemu yize, ikoreshwa na data.
Ni ubuhe buryo bwo gutakaza butuma DeepSpeech yitoza adafite urwego-urwego ruhuza amajwi ninyandiko?
CTC itangiza ikimenyetso cyuzuye kandi igiteranyo hejuru yingirakamaro zose zisenyuka zigwa kumyandiko yagenewe, ikuraho ibikenewe kuri buri kirango.
Ni ubuhe butumwa bukuru bwububiko DeepSpeech yamamaje?
DeepSpeech yasimbuye umuyoboro gakondo wibyiciro byinshi hamwe numuyoboro umwe wimyitozo yatojwe kurangiza, impinduka nini mubishushanyo bya ASR.
Kuki DeepSpeech ikoresha ibyerekezo byombi bigaruka?
Ibyerekezo bibiri RNN itunganya urukurikirane mubyerekezo byombi, bityo buri gisohoka cyungukirwa nibijyanye na acoustic, bikosora neza.
Ni ubuhe bwoko bwo kwinjiza ibintu bisanzwe DeepSpeech ikora?
Icyitegererezo gikoresha ikadiri-urwego rwamajwi nka spekrogramu cyangwa MFCCs kandi ikanasohora imiterere ishoboka kuri buri gihe cyintambwe.
Niki kongerwaho kenshi mugihe cya decode kugirango tunonosore ijambo rya DeepSpeech guhitamo hamwe nimyandikire?
Guteranya ibyasohotse muri acoustic hamwe nururimi rwo hanze mugihe cya decoding bifasha sisitemu kubyara amagambo menshi yimyandikire.