Ubuyobozi bwa Audio AI

Jasper na QuartzNet ASR

Jasper na QuartzNet nicyitegererezo cya NVIDIA cyerekana imenyekanisha ryamagambo, hamwe na QuartzNet yari ntoya cyane, yongeye gushushanya Jasper.

2 min somaIbiherutse kuvugururwa

Incamake

They matter for showing how to get strong accuracy with far fewer parameters, ideal for deployment.

Kwibira cyane

Jasper (Just Another Speech Recognizer), yasohowe na NVIDIA muri 2019, ni umuyoboro wimbitse wa 1D, ugera ku byiciro 54, ushushanya mel-spectrogram ibiranga inyuguti ukoresheje igihombo cya CTC. Yatangije imiyoboro yuzuye isigaye kuburyo gradients zitemba neza zinyuze mubice byimbitse. QuartzNet, yasohotse muri uwo mwaka, yagumanye imiterere ya Jasper ariko isimbuza imyanzuro isanzwe hamwe nigihe-umuyoboro utandukanijwe, igabanya buri muyunguruzi mo imyanzuro yigihe gito hamwe numuyoboro uhuza intambwe. Iyi ngingo yagabanije ibipimo kuva kuri Jasper hafi miliyoni 333 kugeza kuri miliyoni 19 mugihe bihuye neza na Librispeech. Ubwato bwombi mubikoresho bya NVIDIA bya NeMo kandi byateguwe kugirango byihute byihuta bya GPU no gufata umwanzuro-nyabyo, bituma biba ibyamamare byubaka ASR.

Ubushishozi

Imikorere ya QuartzNet iva mugihe-umuyoboro utandukanya imyanzuro, igitekerezo kimwe inyuma ya MobileNet. Ubusanzwe 1D imyanzuro ivanga umwanya numuyoboro hamwe, bigura K inshuro C-mubihe C-out uburemere. Kubitandukanya muburyo bwimbitse mugihe cyongeweho hiyongereyeho 1x1 guhuza ibitekerezo kumiyoboro bigabanya ibipimo kugeza K inshuro C wongeyeho C-mubihe C-hanze. Bishyizwe mubice bisigaye kandi byahuguwe hamwe na CTC, ibi bitanga hafi-Jasper neza kubice byubunini bwikitegererezo na compte.

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.

Kazoza ka Jasper na QuartzNet ASR

Imirongo ya QuartzNet itandukanijwe-yemeza yatumye yerekeza kuri Citrinet ya NVIDIA hamwe na moderi ya Conformer ikoreshwa cyane, yongeraho kwiyitaho kugirango ifate imiterere yisi yose hamwe n’imyanzuro yaho. Tegereza gukomeza urugendo rugana kuri Hybrid convolution-hongeweho-kwitondera imyubakire hamwe na transducer (RNN-T) decoders yo gutemba. Isomo ryibanze, ibipimo-bikora neza kubijyanye nigihe cyoherejwe nigihe gikwiye, bikomeza kuba hagati nkuko ASR isunika kuri terefone, imodoka, nibikoresho byashizwemo.

Gushyira mu bikorwa Isi

Kwandika-igihe-hamwe nabafasha mu majwi boherejwe kuri NVIDIA GPUs ukoresheje ibikoresho bya NeMo

Hindura kandi ushiremo ASR aho QuartzNet ntoya yintambwe ihuye nibikoresho bibitse

Kugenzura neza kwipimisha QuartzNet kugenzura kuri domaine yihariye yamagambo nkubuvuzi cyangwa amategeko

Ihamagarwa-ryisesengura ryandika amajwi manini byihuse kandi bidahenze

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

1

Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.

2

Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.

3

Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.

4

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 Jasper and QuartzNet ASR quiz

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

Tangira ikibazo

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

Ubuyobozi bukurikira

Wav2Inyandiko Yemeza ASR

Ibibazo bikunze kubazwa

What is Jasper and QuartzNet ASR?

Jasper na QuartzNet nicyitegererezo cya NVIDIA cyerekana imenyekanisha ryamagambo, hamwe na QuartzNet yari ntoya cyane, yongeye gushushanya Jasper. Bifite akamaro ko kwerekana uburyo bwo kubona ukuri gukomeye hamwe nibintu bike cyane, byiza kubyoherejwe.

Ni ubuhe bushya bw'ingenzi butuma QuartzNet ikoresha ibipimo bike ugereranije na Jasper?

QuartzNet isimbuza imyanzuro isanzwe hamwe nigihe-umuyoboro utandukanya imyanzuro, igabanya cyane ibipimo bibarwa mugihe ubitse neza.

Hafi ya QuartzNet ibipimo bingahe ugereranije na miliyoni 333 za Jasper?

QuartzNet igabanuka kugera kuri miriyoni 19, hafi 17x yagabanutse, mugihe ihuye na Librispeech ya Jasper.

Niyihe sosiyete yateje imbere Jasper na QuartzNet?

Izi moderi zombi zakozwe na NVIDIA kandi zitangwa binyuze muri NeMo yo kuganira kwa AI.

Ni ubuhe buryo bwo gutakaza Jasper na QuartzNet bakoresha mu kwitoza badahuje neza?

Byombi ukoreshe igihombo CTC, ihuza amajwi-ihindagurika ryamajwi hamwe nimiterere ikurikirana udakeneye ibirango kumurongo.

Ni ubuhe buryo bwubaka bufasha gradients kunyura mumurongo wimbitse wa Jasper (kugeza kuri 54-layer)?

Jasper ikoresha imiyoboro isigaye (skip) kugirango gradients ikwirakwiza neza binyuze mumurongo wimbitse.