Jasper da QuartzNet ASR
Jasper da QuartzNet sune samfurin gane magana na ƙarshe-zuwa-ƙarshe na NVIDIA, tare da QuartzNet kasancewar ƙaramin ƙarami, ingantaccen tsarin Jasper.
Dubawa
They matter for showing how to get strong accuracy with far fewer parameters, ideal for deployment.
Zurfafa nutsewa
Jasper (Wani Mai Gane Magana kawai), wanda NVIDIA ta saki a cikin 2019, cibiyar sadarwa ce mai zurfi ta 1D, har zuwa yadudduka 54, wanda ke tsara fasalin mel-spectrogram zuwa haruffa ta amfani da asarar CTC. Ya gabatar da raƙuman haɗin kai don haka gradients suna gudana cikin tsafta ta cikin tari mai zurfi sosai. QuartzNet, wanda aka saki a wannan shekarar, ya kiyaye tsarin toshewar Jasper amma ya maye gurbin daidaitattun juzu'ai tare da rabe-raben tashoshi na lokaci, yana raba kowane tacewa zuwa juzu'i mai zurfi na ɗan lokaci da matakin haɗakar tashoshi mai ma'ana. Wannan ƙididdiga ya rage ma'auni daga Jasper kusan miliyan 333 zuwa kusan miliyan 19 yayin da ya dace da daidaito akan Librispeech. Dukansu suna jigilar kaya a cikin kayan aikin NeMo na NVIDIA kuma ana sauraron su don horarwar GPU cikin sauri da kuma ainihin lokacin, yana mai da su shahararrun tubalan ginin don samarwa ASR.
Fahimtar Fasaha
Ingancin QuartzNet ya fito ne daga rikice-rikicen tashoshi na lokaci-lokaci, ra'ayi iri ɗaya a bayan MobileNet. Juyin juya hali na 1D na yau da kullun yana haɗa lokaci da tashoshi tare, yana biyan K sau C-in lokutan C-fita nauyi. Raba shi cikin juzu'i mai zurfi akan lokaci tare da juzu'i mai ma'ana 1x1 akan tashoshi yana rage sigogi zuwa K sau C da C-in sau C-out. An tattara shi a cikin ragowar tubalan kuma an horar da shi tare da CTC, wannan yana ba da daidaito kusa-Jasper a ɗan ƙaramin girman samfurin da ƙididdigewa.
Dabarun Tasiri
Shiga ku isa
Yana inganta samun dama ta hanyar rubutu, ba da labari, da mu'amalar murya.
Kudin da kasafin kuɗi
Ƙungiyoyin kafofin watsa labaru na iya jigilar sauti mai gogewa cikin sauri tare da ƙaramin kasafin kuɗi.
Gudu da sikelin
Tsarin fuskantar abokin ciniki na iya aiwatar da hulɗar magana a mafi girman ma'auni.
Makomar Jasper da QuartzNet ASR
Rabe-raben juzu'i na QuartzNet ya jagoranci kai tsaye zuwa Citrinet na NVIDIA da samfuran Conformer da ake amfani da su sosai, waɗanda ke ƙara hankalin kai don ɗaukar mahallin duniya tare da rikice-rikice na gida. Yi tsammanin ci gaba da motsi zuwa ga haɗaɗɗun juzu'i-da-hankali gine-gine da masu gyarawa (RNN-T) don yawo. Babban darasi, ingantaccen juzu'i don ƙwanƙwasa da turawa na ainihi, ya kasance tsakiya yayin da ASR ke turawa kan wayoyi, motoci, da na'urori masu haɗawa.
Aiwatar da Gaskiyar Duniya
Rubutun lokaci na ainihi da mataimakan murya da aka tura akan NVIDIA GPUs ta kayan aikin NeMo
Edge da saka ASR inda ƙaramin sawun QuartzNet ya dace da na'urori masu ƙuntata ƙwaƙwalwar ajiya
Ingantattun wuraren bincike na QuartzNet da aka riga aka horar don takamaiman ƙamus na yanki kamar likitanci ko sharuddan doka
Ƙididdigar cibiyar kira tana rubuta manyan kundin sauti cikin sauri da farashi mai inganci
Hatsari & Tsare-tsare
Rashin amfani da murya da haɗarin kwaikwaya yana ƙaruwa lokacin da aka rasa izini.
Daidaituwa na iya faɗuwa cikin lafuzza, yaruka, ko mahalli masu hayaniya.
Ana iya kuskuren sauti na roba don ingantacciyar magana ba tare da bayyananniyar lakabi ba.
Taswirar Hanya
Sami tabbataccen izini don ɗaukar murya, cloning, da sake amfani.
Gwajin ingantattun masu magana daban-daban da yanayin baya.
Ƙayyade lokacin da dole ne ɗan adam ya duba ko ya amince da abubuwan da aka fitar.
Yi lakabin sauti na roba da kuma adana bayanan da aka tabbatar don yin lissafi.
Ci gaba da Bincike
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.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Jagora na gaba
Wav2Letter Convolutional ASR
Tambayoyin da ake yawan yi
What is Jasper and QuartzNet ASR?
Jasper da QuartzNet sune samfurin gane magana na ƙarshe-zuwa-ƙarshe na NVIDIA, tare da QuartzNet kasancewar ƙaramin ƙarami, ingantaccen tsarin Jasper. Suna da mahimmanci don nuna yadda ake samun daidaito mai ƙarfi tare da ƴan sigogi masu nisa, manufa don turawa.
Wane maɓalli na ƙididdigewa ne ke ba wa QuartzNet damar amfani da ƙananan sigogi fiye da Jasper?
QuartzNet yana maye gurbin daidaitattun juzu'ai tare da rabe-raben rabe-raben tashoshi na lokaci, da yanke ƙidayar siga yayin kiyaye daidaito.
Kusan sigogi nawa ne QuartzNet ke da shi idan aka kwatanta da na ~333 na Jasper?
QuartzNet yana raguwa zuwa kusan sigogi miliyan 19, kusan raguwar 17x, yayin da ya dace da daidaiton Librispeech na Jasper.
Wane kamfani ne ya haɓaka duka Jasper da QuartzNet?
Duk samfuran NVIDIA ne suka haɓaka kuma ana rarraba su ta kayan aikin NeMo na AI na tattaunawa.
Wane aikin hasara ne Jasper da QuartzNet ke amfani da su don horarwa ba tare da daidaitawa ba?
Dukansu suna amfani da asarar CTC, wanda ke daidaita sauti mai tsayi zuwa jerin haruffa ba tare da buƙatar kowane lakabin firam ba.
Wanne fasalin tsarin ne ke taimaka wa gradients su gudana ta hanyar sadarwar Jasper mai zurfi (har zuwa Layer 54)?
Jasper yana amfani da haɗe-haɗe masu yawa (tsalle) don haka gradients suna yaduwa da tsabta ta cikin zurfin juzu'in tari.