Ngwa njirimara Okwu Kaldi
Kaldi bụ ngwa n'efu, mepere emepe nke ghọrọ usoro nyocha kachasị maka iwulite sistemụ njirimara okwu.
Nchịkọta
It matters because for nearly a decade it was the go-to foundation for academic and industrial ASR work.
Ime miri emi
Edere Kaldi, nke ewepụtara na 2011 ma Daniel Povey duziri, na C ++ nwere usoro nri jikọtara ọnụ site na bash na scripts Perl. O wuru n'elu pipeline ASR nke kpochapụwo: wepụta njirimara egwu (MFCCs ma ọ bụ filterbanks), ụda phoneme na ụdị Gaussian Mixture Models ma ọ bụ, emesịa, netwọk akwara miri emi, wee jikọta ụdị ụda olu, lexicon ịkpọ okwu, na ụdị asụsụ n'ime otu eserese a na-achọgharị. Nhọrọ teknuzu ya na-akọwapụta bụ iji ndị ntụgharị steeti dị oke arọ (WFSTs) sitere na ọbaakwụkwọ OpenFST iji kesaa isi mmalite ihe ọmụma n'otu eserese ngbanwe. Kaldi butere 'ntụziaka' maka ntọala data ọkọlọtọ dị ka Switchboard, Librispeech, na Wall Street Journal, na-ahapụ ndị nyocha ka ha mepụtaghachi nsonaazụ ọgbara ọhụrụ. Ọ ghọrọ mmejuputa nrụtụ aka nke edobere sistemu ọhụrụ.
Nghọta nka nka
Isi aghụghọ Kaldi na-edepụta WFST anọ n'ime otu eserese a na-akpọ HCLG: H maps neural-net ma ọ bụ GMM na-ekwu na ekwentị dabere na ọnọdụ, C na-ejikwa ọnọdụ ụda olu (triphones), L bụ okwu okwu lexicon na-ese ekwentị na okwu, na G bụ ụdị asụsụ. Ịba ụba transducers ndị a na ịkwalite nsonaazụ ya na-ewepụta otu eserese nke onye nrụpụta na-eji Viterbi algọridim beam-pruned, na-atụgharị fremu ọdịyo ka ọ bụrụ usoro okwu nwere ike ịrụ ọrụ nke ọma.
Mmetụta atụmatụ
Nweta na iru
Ọ na-eme ka nnweta ya dịkwuo mma site na ndegharị, ịkọ akụkọ, na ntụgharị olu.
Ọnụ ego na mmefu ego
Ndị otu mgbasa ozi nwere ike ibubata ọdịyo a na-egbu maramara ngwa ngwa site na iji obere mmefu ego.
Ọsọ na ọnụ ọgụgụ
Sistemụ na-eche ihu ndị ahịa nwere ike hazie mkparịta ụka n'ọtụtụ buru ibu.
Ngwa ihe nchọpụta okwu Kaldi ga-eme n'ọdịnihu
Usoro HMM-DNN ngwakọ Kaldi ejiriwo ụdị akwara ngwụcha ruo na ngwụcha na-esetịpụ ọdịyo ozugbo na ederede. Ihe oru ngo nke ga-anọchi anya Daniel Povey, k2 (ya na Icefall na Lhotse gburugburu ebe obibi), na-atụgharị echiche Kaldi's WFST na PyTorch nwere automata nwere oke steeti. Na-atụ anya Kaldi n'onwe ya ka ọ ga-abụ ntụaka akụkọ ihe mere eme yana ngwa nkuzi, ebe ụmụ ya na-eche echiche na-ejikọta ngbanwe oge gboo a haziri ahazi yana ụdị ụda mgbako ọgbara ọhụrụ na nke na-ahụ maka onwe ya.
Mmejuputa n'ezie n'ụwa
Ụlọ nyocha ụlọ akwụkwọ na-emepụtagharị Librispeech na Switchboard benchmarks iji kwado nyocha ihe ngosi nka ọhụrụ.
Iwulite sistemu iwu olu omenala maka asụsụ ndị dị ala ma ọ bụ obere asụsụ site na iji usoro nri Kaldi
Nhazi nke ọdịyo manyere na ndegharị maka asụsụ, imepụta dataset, na oge ndepụta okwu.
Ịkwado nchọ olu n'oge na nkwuwa okwu na-azụ azụ n'ụlọ ọrụ tupu ụdị ngwụcha ruo ọgwụgwụ etozuo
Ihe ize ndụ & okporo ụzọ nche
Iji olu eme ihe na ihe egwu mpụta ga-abawanye mgbe nkwenye na-efu.
Izi ezi nwere ike ịdaba n'ofe ụda olu, olumba ma ọ bụ gburugburu mkpọtụ.
Enwere ike imehie ọdịyo sịntetik dị ka ezigbo okwu na-enweghị akara doro anya.
Map mmejuputa
Nweta nkwenye doro anya maka ijide olu, imechi, na ijigharị.
Nwale ogo n'ofe ndị na-ekwu okwu dị iche iche yana ọnọdụ ndabere.
Kọwaa mgbe mmadụ ga-enyocha ma ọ bụ kwado nsonye.
Deba aha ọdịyo sịntetik ma debe ndekọ ihe ndekọ maka ịza ajụjụ.
Nọgide na-eme nchọpụta
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
Ntuziaka na-esote
Mmata Okwu ntanye
Ajụjụ a na-ajụkarị
What is Kaldi Speech Recognition Toolkit?
Kaldi bụ ngwa n'efu, mepere emepe nke ghọrọ usoro nyocha kachasị maka iwulite sistemụ njirimara okwu. Ọ dị mkpa n'ihi na ihe fọrọ nke nta ka ọ bụrụ afọ iri ọ bụ ntọala maka agụmakwụkwọ na ụlọ ọrụ ASR.
Kedu asụsụ mmemme ka e dere isi Kaldi?
Edere isi Kaldi na C++ maka ịrụ ọrụ, yana scripts bash na Perl na-ahazi ọzụzụ na nhazi nhazi.
Kedu usoro mgbakọ na mwepụ Kaldi na-eji ijikọ ụdị ụda olu ya, akwụkwọ ọkọwa okwu na ụdị asụsụ ya n'otu eserese ngbanwe?
Kaldi na-edepụta WFST si n'ọbá akwụkwọ OpenFST n'ime otu eserese HCLG nke onye nrụpụta na-achọ.
Kedu onye bụ isi onye okike na onye ndu nke ọrụ Kaldi?
Daniel Povey duziri mmepe nke Kaldi, nke mbụ ewepụtara na 2011, ma mesịa malite ọrụ onye nọchiri k2.
N'ime pipeline kpochapụwo nke Kaldi, kedu ihe GMMs (Gaussian Mixture Models) na-ejibu eme ihe?
Ndị GMM chepụtara ohere ụda ụda nke steeti phoneme tupu netwọk akwara dị omimi dochie ha na sistemu ngwakọ.
Gịnị bụ aha usoro gburugburu ebe obibi dabere na PyTorch ọgbara ọhụrụ nye Kaldi?
k2, yana Icefall na Lhotse, na-emeghachite echiche steeti njedebe Kaldi n'ụdị dị iche, PyTorch-enyi na enyi.