Ntuziaka Audio AI

Jasper na QuartzNet ASR

Jasper na QuartzNet bụ NVIDIA's end-na-end convolutional convolutional speech model, na QuartzNet bụ obere obere, nhazigharị nke ọma nke Jasper.

2 nkeji na-agụEmelitere ikpeazụ

Nchịkọta

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

Ime miri emi

Jasper (Nanị Ihe Nchọpụta Okwu Ọzọ), nke NVIDIA weputara na 2019, bụ netwọọdụ convolutional 1D miri emi, ruo ọkwa 54, nke na-esepụta njirimara mel-spectrogram na mkpụrụedemede na-eji mfu CTC. Ọ webatara njikọ dị nhịahụ ka gradients na-asọpụta nke ọma site na ngwugwu miri emi. QuartzNet, nke ewepụtara n'otu afọ ahụ, debere usoro mgbochi Jasper mana dochie convolutions ọkọlọtọ na ngbanwe nke ọwa oge, na-ekewa nzacha nke ọ bụla n'ime ntụgharị oge omimi miri emi yana usoro ịgwakọta ọwa dabara adaba. Nrụpụta a wepụrụ paramita sitere na ihe dị ka nde 333 nke Jasper gbadara nde 19 ka ọ dabara na Librispeech. Abụọ ụgbọ mmiri na NVIDIA's NeMo Toolkit ma na-ege ntị maka ọzụzụ GPU ngwa ngwa na ntinye oge, na-eme ka ha bụrụ ihe ewu ewu ewu maka mmepụta ASR.

Nghọta nka nka

Ịrụ ọrụ nke QuartzNet na-abịa site na ọgba aghara nkewa nkewa oge, otu echiche dị n'azụ MobileNet. Mgbagha 1D nkịtị na-agwakọta oge na ọwa ọnụ, na-eri K ugboro C-in ugboro C-out arọ. Ikewapụ ya n'ime mgbagwoju anya miri emi n'ime oge yana 1x1 ntụgharị uche na ọwa na-ebelata paramita na K ugboro C gbakwunyere C-in times C-out. Ejikọtara ya na ihe ndị fọdụrụ ma jiri CTC zụọ ya, nke a na-enye ihe ziri ezi nso-Jasper n'obere ntakịrị nke nha ihe nlereanya yana gbakọọ.

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 Jasper na QuartzNet ASR

Usoro nkewapụ-convolution nke QuartzNet dugara ozugbo na NVIDIA's Citrinet na ụdị Conformer a na-ejikarị eme ihe, nke na-agbakwunye nlebara anya onwe ya iji weghara ọnọdụ ụwa n'akụkụ mgbagha mpaghara. Na-atụ anya ka aga n'ihu na-aga n'ihu n'ebe ngwakọ convolution-gbakwunyere-nleba anya architectures na transducer (RNN-T) decoders maka nkwanye. Ihe nkuzi a bụ isi, ngbanwe na-arụ ọrụ nke ọma maka ihu na mbugharị oge, ka bụ etiti ka ASR na-agbanye na ekwentị, ụgbọ ala, na ngwaọrụ agbakwunyere.

Mmejuputa n'ezie n'ụwa

Ndegharị oge na ndị enyemaka olu ebugara na NVIDIA GPU site na ngwa ngwa NeMo

Edge na agbakwunyere ASR ebe obere akara ụkwụ QuartzNet dabara na ngwaọrụ ndị nwere nsogbu ebe nchekwa

Nleba anya QuartzNet emezigharịrị nke ọma maka mkpụrụokwu akọwapụtara ngalaba dịka ọgwụ ma ọ bụ usoro iwu.

Nchịkọta etiti oku na-edegharị nnukwu olu ngwa ngwa yana ọnụ ahịa dị irè

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 Jasper and QuartzNet ASR 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

Wav2Letter Convolutional ASR

Ajụjụ a na-ajụkarị

What is Jasper and QuartzNet ASR?

Jasper na QuartzNet bụ NVIDIA's end-na-end convolutional convolutional speech model, na QuartzNet bụ obere obere, nhazigharị nke ọma nke Jasper. Ha dị mkpa maka igosi otu esi enweta izi ezi siri ike na parampat dị ole na ole, nke dị mma maka ibugharị.

Kedu isi ihe ọhụrụ na-ahapụ QuartzNet jiri parampat dị obere karịa Jasper?

QuartzNet dochie convolutions ọkọlọtọ na ọwa oge nkewa, na-egbutu ọnụ ọgụgụ oke ma na-echekwa izi ezi.

Ihe dị ka nkeji ole QuartzNet nwere atụnyere nde ~333 nke Jasper?

QuartzNet na-agbada ruo ihe dị ka nde 19, ihe dị ka mbelata 17x, ebe ọ dabara na izi ezi Librispeech Jasper.

Kedu ụlọ ọrụ mepụtara ma Jasper na QuartzNet?

NVIDIA mepụtara ụdị abụọ a wee kesaa ya site na ngwa mkparịta ụka NeMoal AI ya.

Kedu ọrụ mfu Jasper na QuartzNet na-eji na-azụ na-enweghị nhazi doro anya?

Ha abụọ na-eji mfu CTC, nke na-ahazi ọdịyo ogologo ogologo na usoro agwa na-achọghị akara ọ bụla.

Kedu njiri arụ ọrụ na-enyere gradients na-agafe na netwọk Jasper dị omimi (ruo 54-layer)?

Jasper na-eji njikọ ndị fọdụrụ (ịwụ elu) dị ukwuu ka gradients na-agbasa nke ọma site na nchịkọta mgbagwoju anya miri emi.