Igikoresho cyo Kumenyekanisha Kaldi
Kaldi nigitabo cyubusa, gifungura-isoko-ibikoresho byahindutse urubuga rwubushakashatsi bwubaka sisitemu yo kumenya imvugo.
Incamake
It matters because for nearly a decade it was the go-to foundation for academic and industrial ASR work.
Kwibira cyane
Kaldi, yasohotse mu 2011 kandi iyobowe na Daniel Povey, yanditswe muri C ++ hamwe na resept zifatanije hamwe na bash na Perl. Yubatswe kumuyoboro wa kera wa ASR: gukuramo ibintu bya acoustic (MFCCs cyangwa filterbanks), amajwi ya foneme yerekana hamwe na Gaussian Mixture Models cyangwa, nyuma, imiyoboro yimbitse y’imitsi, kandi igahuza icyitegererezo cya acoustic, imvugo yamagambo, hamwe nururimi rwindimi mubishushanyo bimwe byashakishwa. Ihitamo ryayo rya tekiniki ryakoreshaga transducers iremereye (WFSTs) ivuye mubitabo bya OpenFST kugirango ihimbe ubumenyi bwose mubishushanyo mbonera. Kaldi yohereje 'resept' kuri dataseti zisanzwe nka Switchboard, Librispeech, na Wall Street Journal, bituma abashakashatsi berekana ibisubizo bigezweho. Byahindutse ishyirwa mubikorwa sisitemu nshya yashizweho.
Ubushishozi
Amayeri yibanze ya Kaldi ni uguhimba WFST enye mu gishushanyo kimwe cyitwa HCLG: H ikarita ya neural-net cyangwa GMM ivuga kuri terefone zishingiye ku miterere, C ikora imiterere ya fonetike (triphones), L ni imvugo yerekana ikarita ya terefone ku magambo, naho G ni urugero rwururimi. Kugwiza izo transducers no guhitamo ibisubizo bitanga igishushanyo kimwe decoder ishakisha hamwe na algorithm ya Viterbi yacishijwe bugufi, ihindura amajwi kumajwi muburyo bukurikirana ijambo.
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 ha Kaldi Imvugo yo Kumenyekanisha
Kaldi ya Hybrid ya HMM-DNN yasimbujwe ahanini na moderi yimitsi iherezo-iherezo yerekana ikarita yerekana amajwi kumyandiko. Umushinga uzasimbura Daniel Povey, k2 (hamwe na ecosystem ya Icefall na Lhotse), yongeye gutekereza ibitekerezo bya WFST ya Kaldi muri PyTorch hamwe na automatique itandukanye ya leta. Witege ko Kaldi ubwayo izakomeza kuba amateka nigikoresho cyo kwigisha, mugihe ababakomokaho bahuza decoding ya classique yubatswe hamwe na moderi igezweho kandi ishingiye kuri acoustic.
Gushyira mu bikorwa Isi
Laboratwari yamasomo yerekana ibipimo ngenderwaho bya Librispeech na Switchboard kugirango yemeze ubushakashatsi bushya bwo kwerekana imiterere ya acoustic
Kubaka amajwi yigenga ya sisitemu ya sisitemu yo hasi cyangwa indimi nkeya ukoresheje Kaldi resept
Guhatira guhuza amajwi inyandiko-mvugo ku bumenyi bw'indimi, kurema dataset, hamwe na subtitle igihe
Imbaraga zo gushakisha amajwi hakiri kare hamwe nigitekerezo gisubira inyuma munganda mbere yuko impera zanyuma zirangira
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 Kaldi Speech Recognition Toolkit 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
Kwongorera kuvuga
Ibibazo bikunze kubazwa
What is Kaldi Speech Recognition Toolkit?
Kaldi nigitabo cyubusa, gifungura-isoko-ibikoresho byahindutse urubuga rwubushakashatsi bwubaka sisitemu yo kumenya imvugo. Nibyingenzi kuko mumyaka hafi icumi aribwo buryo bwo gushingira kumirimo ya ASR yamasomo ninganda.
Ni uruhe rurimi rwo gutangiza porogaramu Kaldi yanditsemo?
Kaldi yibanze yanditswe muri C ++ kugirango ikore, hamwe na bash na Perl inyandiko zitegura amahugurwa hamwe na decoding resept.
Ni ubuhe buryo bw'imibare Kaldi akoresha kugira ngo ahuze imiterere ya acoustic, lexicon, hamwe nururimi rwindimi mu gishushanyo kimwe?
Kaldi yahimbye WFST kuva mubitabo bya OpenFST mubishushanyo bimwe bya HCLG decoder ishakisha.
Ninde waremye mbere nuyobora umushinga wa Kaldi?
Daniel Povey yayoboye iterambere rya Kaldi, yasohotse bwa mbere muri 2011, nyuma atangira umushinga uzasimbura k2.
Mu miyoboro ya kera ya Kaldi, ni ubuhe GMMs (Moderi ya Gaussian)?
GMMs yerekanye uburyo bwa acoustic bushoboka bwa foneme leta mbere yuko imiyoboro yimbaraga nini yabasimbuye muri sisitemu ya Hybrid.
Izina rya kijyambere rya PyTorch rishingiye ku gusimbuza urusobe rwa Kaldi ni irihe?
k2, hamwe na Icefall na Lhotse, byongera ibitekerezo bya Kaldi bya leta-bitagira ingano muburyo butandukanye, bushimishije PyTorch.