DeepSpeech Architecture
DeepSpeech waa qaabka aqoonsiga hadalka dhamaadka-ilaa-dhamaadka oo ay soo bandhigtay Baidu 2014 kaas oo khariidado sifooyin maqal cayriin ah si toos ah ugu dhejiya qoraalka iyada oo la adeegsanayo shabakad neural ah oo soo noqnoqota oo lagu tababaray khasaaraha CTC.
Dulmar
It helped pioneer the shift away from complex, hand-engineered ASR pipelines toward learned, data-driven systems.
quusid qoto dheer
Aqoonta hadalka qadiimiga ah waxay isku toleen moodooyin dhawaaq gaar ah, qaamuusyo ku dhawaaqida, iyo moodooyinka luqadda oo leh qaybo gacanta lagu hagaajiyay. DeepSpeech ayaa taas inteeda badan ku beddeshay hal shabakad oo neerfaha ah oo u tababaran dhammaad ilaa dhammaad. Nashqaddeedu waxa ay qaadaa spectrogram ama MMFC muuqaalada fiimyada maqalka ee gaagaaban oo ay ku quudiyaan dhawr lakab oo si buuxda isugu xidhan, lakab laba jiho ah oo soo noqnoqda kaas oo qabta macnaha hore iyo mustaqbalka, iyo lakab wax soo saar ah oo soo saara qaybinta suurtogalka ah ee jilayaasha talaabo kasta. Muhiimad ahaan, waxay isticmaashaa Kala-saar ku-meel-gaar ah oo ku-meel-gaar ah (CTC), taas oo u oggolaanaysa shabakadu inay bartaan isku-habaynta u dhexeeya maqalka iyo qoraalka iyada oo aan loo baahnayn calaamado heer-jir ah. Mozilla ayaa markii dambe soo saartay hirgelinta il furan oo caan ah (oo leh noocyo cusub oo isticmaalaya LSTM-ku-saleysan, nashqada la socon karo), taasoo ka dhigaysa habka si ballaaran loo heli karo.
Aragtida Farsamada
Awood bixiyaha furaha waa khasaaraha CTC. Hadalka iyo qoraalku maaha kuwo isku toosan oo qaab-jir ah, markaa CTC waxay soo bandhigaysaa calaamad 'madhan' oo waxay koobaysaa dhammaan toosinta suurtogalka ah ee ku burburta qoraalka bartilmaameedka. Tani waxay u sahlaysaa moodelku inuu soo saaro jile mar kasta oo uu barto halka dhawaaqyada khariidadda xarfaha si toos ah u socdaan. RNN laba jiho ah ayaa siinaysa saadaal kasta marin u helida macnaha dhawaaqa ku xeeran, iyo qaabka luqadda n-gramka dibadda ayaa inta badan lagu daraa wakhtiga go'aaminta si loo hagaajiyo higaada iyo doorashada kelmadda.
Saamaynta Istiraatijiyadeed
Helitaanka iyo gaarsiinta
Waxay wanaajisaa marin u helida iyada oo loo marayo qoraal-qorid, sheeko, iyo is-dhexgalyo cod.
Qiimaha iyo miisaaniyada
Kooxaha warbaahintu waxay ku soo rari karaan codka sifaysan si degdeg ah iyagoo wata miisaaniyado yaryar.
Xawaaraha iyo miisaanka
Nidaamyada u jeedda macmiisha waxay ka baaraandegi karaan isdhexgalka hadalka si weyn.
Mustaqbalka DeepSpeech Architecture
DeepSpeech lafteedu waxa si weyn u beddelay feejignaan- iyo qaab-dhismeedka ku-salaysan beddelka (Conformer, Whisper, wav2vec 2.0) kaas oo qabsada macnaha guud iyo is-maamulida maqalka aan la summadaysan. Laakiin fikradaheeda asaasiga ah, tababarka dhamaadka-ilaa-dhamaadka iyo CTC-dajinta, ayaa weli ah aasaaska waxayna wali ka dhex muuqdaan nidaamyada isku-dhafan ee casriga ah. Dhaxalku waa fikradeed: waxa ay caddeeyeen in hal nooc oo la bartay uu la tartami karo dhuumaha si weyn loo farsameeyey, taas oo u gogol xaaraysa moodooyinka aasaasiga ah ee hadalka ee waaweyn, luuqado badan ku hadla, is-maamula maanta.
Dhaqangelinta Adduunka-dhabta ah
Qad la'aan, aqoonsiga amarka codka aaladda ee codsiyada diirada saaraya sirta ah iyadoo la adeegsanayo Mozilla's DeepSpeech furan
Soo saarida qoraalada qoraalka ah ee podcasts ama muxaadarooyinka adiga oo aan ku tiirsanayn adeegga daruuraha
Barida aasaaska dhamaadka-ilaa-dhamaadka ASR iyo khasaaraha CTC ee koorsooyinka barashada mashiinka-jaamacadda
Dhisidda isku-xidhka codka gaarka ah ee IoT ama aaladaha ku-xidhan halka loo baahan yahay garane fudud oo la socon karo
Khatarta & Dariiqyada Ilaalada
Si xun u isticmaalka codka iyo khataraha is-yeelyeelku way kordhaan marka oggolaanshaha la waayo.
Saxnimadu waxay hoos ugu dhici kartaa lahjadaha, lahjadaha, ama jawiga buuqa badan.
Maqalka synthetic waxaa lagu khaldi karaa hadal dhab ah iyada oo aan si cad loo calaamadin.
Qorshe Hawleedka Dhaqangelinta
Hel ogolaansho cad oo ku saabsan qabashada codka, xidhitaanka, iyo dib u isticmaalka
Tijaabi tayada ku hadasha kala duwan iyo xaaladaha asalka.
Qeex marka bani'aadamku ay tahay inuu dib u eego ama oggolaado wax soo saarka.
Ku calaamadee codka synthetic oo xafid diiwaannada la-xisaabtanka.
Sii wad Sahaminta
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
Hagaha xiga
Qaab-dhismeedka Conformer
Su'aalaha soo noqnoqda
What is DeepSpeech Architecture?
DeepSpeech waa qaabka aqoonsiga hadalka dhamaadka-ilaa-dhamaadka oo ay soo bandhigtay Baidu 2014 kaas oo khariidado sifooyin maqal cayriin ah si toos ah ugu dhejiya qoraalka iyada oo la adeegsanayo shabakad neural ah oo soo noqnoqota oo lagu tababaray khasaaraha CTC. Waxa ay gacan ka gaysatay in laga guuro kakan, dhuumaha ASR-gacanta lagu farsameeyay ee u jihaystay wax la bartay, habab xogta lagu hago.
Waa maxay shaqada luminta u oggolaanaysa DeepSpeech in ay tababarto iyada oo aan la isku toosin heerka qaab-dhismeedka maqalka iyo qoraalka?
CTC waxay soo bandhigaysaa calaamad madhan oo waxay koobaysaa dhammaan toosinta saxda ah ee ku burburaysa qoraalka la beegsanayo, iyada oo meesha ka saaraysa baahida sumadaha qaab-dhismeedka kasta.
Maxay ahayd falsafada dhismaha ee ugu weyn ee DeepSpeech faafisay?
DeepSpeech waxay ku beddeshay dhuumaha soo jireenka ah ee marxaladaha badan leh hal shabakad neural oo tababaran dhamaadka ilaa dhamaadka, isbeddel weyn oo ku yimid naqshadaynta ASR.
Waa maxay sababta DeepSpeech u isticmaasho lakab laba jiho ah oo soo noqnoqda?
RNN laba jiho ah ayaa isku xigxiga labada jiho, marka wax soo saar kastaa wuxuu ka faa'iideystaa macnaha guud ee dhawaaqa, hagaajinta saxnaanta.
Waa maxay nooca wax gelinta ee uu DeepSpeech caadi ahaan ku shaqeeyo?
Qaabku waxa uu cunaa sifooyinka maqalka heerka-Feemka sida spectrograms ama MMFCs oo soo saara itimaalka jilaha ee talaabo kasta.
Maxaa inta badan lagu daraa wakhtiga go'aaminta si loo hagaajiyo doorashada kelmad DeepSpeech iyo higaada?
Isku-darka wax-soo-saarka cod-bixinta iyo qaabka luqadda dibadda inta lagu jiro kood-samaynta waxay ka caawisaa nidaamku inuu soo saaro erayo iyo higaadyo macquul ah.