Ntuziaka Audio AI

Nkọwapụta maka ịmata okwu

SpecAugment bụ ụzọ dị mfe ma dị ike na-eme ka data dịkwuo ike nke na-ekpuchi ma na-agbagha spectrogram nke okwu iji mee ka ụdị njirimara sie ike karị.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

It boosted accuracy on benchmarks without any new audio or model changes.

Ime miri emi

SpecAugment, nke Google Brain (Park et al.) webatara na 2019, na-akwalite ọzụzụ njiri mara okwu site na-edezi log-mel spectrogram ozugbo karịa ụdị waveform. Ọ na-emetụta atọ arụmọrụ: oge ​​warping, nke ubé agbatị ma ọ bụ compresses audio tinyere oge axis; ugboro ugboro masking, nke zeroes pụọ bandeji nke ugboro ọwa; na nkpuchi oge, nke na-ewepụ ogologo oge usoro. Site n'ịmanye ihe nlereanya ka ọ mata okwu ọbụlagodi mgbe ezoro ezo nke spectrogram, SpecAugment na-arụ ọrụ dị ka nhazigharị ma na-egbochi oke. Ọ dị oke ọnụ ala ma dị irè, na-enyere ụdị ụdị LAS aka iru ọnụ ọgụgụ njehie okwu ọgbara ọhụrụ na LibriSpeech na Switchboard, ọ na-anọgide na-abụ ihe ndabere na pipeline ọzụzụ ASR ọgbara ọhụrụ.

Nghọta nka nka

SpecAugment na-arụ ọrụ na 2D spectrogram dị ka a ga-asị na ọ bụ onyonyo. Nkpuchi ugboro ugboro na-ewepụ ihe mgbochi nke ọwa mel-frequency; oge nkpuchi na-ewepụ ihe mgbochi nke okpokolo agba ugboro ugboro; oge warping na-agbanwe ebe a họọrọ n'akụkụ axis oge site na iji interpolation. Enwere ike itinye ọtụtụ ihe mkpuchi kwa otu okwu. N'ihi na ihe mkpuchi ahụ na-agbanwe oge ọ bụla, ihe nlereanya ahụ na-ahụ nke ọma ọdịiche na-adịghị agwụ agwụ nke ihe atụ ọ bụla, na-emeziwanye mkpokọta na-enweghị ịnakọta data ọhụrụ.

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.

Ọdịnihu nke nkọwapụta maka nnabata okwu

SpecAugment abụrụla ihe ndabara dị nso na njirimara okwu ma na-agbasa na ọrụ ọdịyo ndị ọzọ dị ka nkwenye ọkà okwu na nhazi ụda. Ọrụ ga-eme n'ọdịnihu na-akụ atumatu nkpuchi na-akpaghị aka ma ọ bụ na-emegharị ha n'oge ọzụzụ, ma jikọta spectrogram masking na ebumnobi tupu ọzụzụ ọzụzụ onwe onye na-elekọta. Ka ụdị na-eto eto, mmụba dị ọnụ ala nke na-agbakwunye ike na-enweghị ụda egwu agbakwunyere na-anọgide na-aba uru nke ukwuu, karịsịa maka asụsụ ndị dị ala ebe data dị ụkọ.

Mmejuputa n'ezie n'ụwa

Na-emeziwanye ọnụego mperi okwu na LibriSpeech site na ikpuchi igwe spectrogram n'oge ọzụzụ

Na-ahazi ụdị ASR ngwụcha ruo ọgwụgwụ dị ka LAS ma ọ bụ Conformer iji belata nfefe

Na-emewanye obere dataset maka asụsụ ndị dị ala na-enweghị ndekọ ọdịyo ọhụrụ

Na-emegharị echiche nkpuchi na nkwenye ọkà okwu na nhazi ọkwa mmemme ọdịyo

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

1

Nweta nkwenye doro anya maka ijide olu, imechi, na ijigharị.

2

Nwale ogo n'ofe ndị na-ekwu okwu dị iche iche yana ọnọdụ ndabere.

3

Kọwaa mgbe mmadụ ga-enyocha ma ọ bụ kwado nsonye.

4

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 SpecAugment for Speech Recognition quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

Malite ajụjụ

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ntuziaka na-esote

Ngwa njirimara Okwu Kaldi

Ajụjụ a na-ajụkarị

What is SpecAugment for Speech Recognition?

SpecAugment bụ ụzọ dị mfe ma dị ike na-eme ka data dịkwuo ike nke na-ekpuchi ma na-agbagha spectrogram nke okwu iji mee ka ụdị njirimara sie ike karị. Ọ kwalitere izi ezi na akara akara na-enweghị mgbanwe ọ bụla ọhụrụ ma ọ bụ ihe atụ.

Kedu ihe SpecAugment na-arụ ọrụ?

SpecAugment na-edezi spectrogram (nnọchite anya oge-ugboro) ozugbo karịa ụdị ebili mmiri.

Kedu otu n'ime ọrụ atọ SpecAugment mere?

Arụ ọrụ atọ ahụ bụ ịgbagha oge, nkpuchi ugboro ugboro, na nkpuchi oge.

Gịnị bụ isi nzube nke SpecAugment?

Site na izochi akụkụ nke spectrogram ahụ, ọ na-amanye ihe nlereanya ahụ ka ọ gbasaa, na-ebelata oke mma na iwetu ọnụ ọgụgụ njehie.

Kedu ihe 'time masking' na-eme?

Ikpuchi oge na-ewepụ ihe mgbochi nke okpokolo agba na-esochi n'akụkụ oge nke spectrogram.

Kedu ihe kpatara ịgbanwe ihe mkpuchi n'oge ọ bụla na-enyere aka?

Masks random dị iche iche n'oge ọ bụla pụtara na ihe nlereanya ahụ na-ezute ọdịiche na-adịghị agwụ agwụ, na-emeziwanye mkpokọta na-enweghị data ọhụrụ.