Audio AI JAGORA

DeepSpeech Architecture

DeepSpeech samfurin gane magana ne na ƙarshen-zuwa-ƙarshe wanda Baidu ya gabatar a cikin 2014 wanda ke tsara taswirorin ingantaccen sauti mai jiwuwa kai tsaye zuwa rubutu ta amfani da hanyar sadarwa mai maimaitawa da aka horar tare da asarar CTC.

2 min karatuAn sabunta ta ƙarshe

Dubawa

Ya taimaka majagaba wajen kawar da hadaddun, bututun ASR da aka yi da hannu zuwa ga koyo, tsarin sarrafa bayanai.

Zurfafa nutsewa

Masu gane magana na yau da kullun sun dinke samfuran sauti daban daban, ƙamus na furci, da ƙirar harshe tare da abubuwan da aka gyara da hannu. DeepSpeech ya maye gurbin mafi yawan waccan tare da hanyar sadarwa guda ɗaya da aka horar da ƙarshen zuwa ƙarshe. Gine-ginen nasa yana ɗaukar fasalin sifofi ko MMFC akan gajerun firam ɗin audio kuma yana ciyar da su ta hanyoyin haɗin kai da yawa, madaidaicin madaurin kai wanda ke ɗaukar mahallin daga baya da na gaba, da kuma abin fitarwa yana samar da yiwuwar rarraba kan haruffa a kowane mataki. Mahimmanci, yana amfani da Rarraba Tsari na ɗan lokaci (CTC), wanda ke ba cibiyar sadarwar damar koyon jeri tsakanin sauti da rubutu ba tare da buƙatar alamun matakin firam ba. Daga baya Mozilla ta fitar da sanannen aiwatarwa na buɗe tushen (tare da sabbin nau'ikan ta amfani da tushen LSTM, ƙira mai gudana), yana mai da hanyar samun dama ga kowa.

Fahimtar Fasaha

Maɓalli mai kunnawa shine asarar CTC. Ba a daidaita magana da rubutu ba, don haka CTC ta gabatar da alamar 'blank' tare da ƙididdige duk yuwuwar daidaitawa da ke rugujewa zuwa kwafin da aka yi niyya. Wannan yana bawa samfurin damar fitar da hali kowane mataki lokaci kuma ya koyi inda taswirar sauti zuwa haruffa ta atomatik. RNN na biyu yana ba kowane tsinkaya damar yin amfani da mahallin sauti mai kewaye, kuma ana ƙara ƙirar n-gram na waje a lokacin yanke lamba don inganta rubutun kalmomi da zaɓin kalma.

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 DeepSpeech Architecture

DeepSpeech kanta an maye gurbinsa da hankali- da gine-gine na tushen taswira (Conformer, Whisper, wav2vec 2.0) waɗanda ke ɗaukar tsayin yanayi da kulawa da kai akan sautin da ba a lakafta ba. Amma ainihin ra'ayoyinsa, horo na ƙarshe zuwa ƙarshe da yanke shawara na CTC, sun kasance masu tushe kuma har yanzu suna bayyana a cikin tsarin zamani na zamani. Abin da ya gada yana da ra'ayi: ya tabbatar da cewa samfurin koyo guda ɗaya zai iya yin hamayya da injinan bututun mai, yana ba da hanya ga manyan, harsuna da yawa, ƙirar tushen magana mai kulawa da kai.

Aiwatar da Gaskiyar Duniya

A layi, ƙwarewar muryar murya akan na'urar don aikace-aikacen da aka mayar da hankali kan sirri ta amfani da Buɗe DeepSpeech na Mozilla

Samar da daftarin bayanan kwasfan fayiloli ko laccoci ba tare da dogaro da sabis na girgije ba

Koyar da mahimman abubuwan ASR na ƙarshe zuwa ƙarshe da asarar CTC a cikin darussan koyon injin injin jami'a

Gina mu'amalar murya ta al'ada don IoT ko na'urorin da aka haɗa inda ake buƙatar mai ganewa mai sauƙi, mai rafi

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

1

Sami tabbataccen izini don ɗaukar murya, cloning, da sake amfani.

2

Gwajin ingantattun masu magana daban-daban da yanayin baya.

3

Ƙayyade lokacin da dole ne ɗan adam ya duba ko ya amince da abubuwan da aka fitar.

4

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 DeepSpeech Architecture quiz

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

Fara tambayoyi

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

Jagora na gaba

Conformer Architecture

Tambayoyin da ake yawan yi

Menene DeepSpeech Architecture?

DeepSpeech samfurin gane magana ne na ƙarshen-zuwa-ƙarshe wanda Baidu ya gabatar a cikin 2014 wanda ke tsara taswirorin ingantaccen sauti mai jiwuwa kai tsaye zuwa rubutu ta amfani da hanyar sadarwa mai maimaitawa da aka horar tare da asarar CTC. Ya taimaka majagaba wajen kawar da hadaddun, bututun ASR da aka yi da hannu zuwa ga koyo, tsarin sarrafa bayanai.

Wane aikin asara ya ba DeepSpeech damar horarwa ba tare da daidaita matakin-firam tsakanin sauti da rubutu ba?

CTC yana gabatar da wata alama mara kyau da ƙididdigewa akan duk ingantattun jeri da ke rugujewa zuwa rubutun da aka yi niyya, tare da cire buƙatar alamun kowane-firam.

Menene babban falsafar gine-ginen da DeepSpeech ya shahara?

DeepSpeech ya maye gurbin bututun mai matakai da yawa na gargajiya tare da hanyar sadarwar jijiyoyi ɗaya da aka horar da ƙarshe zuwa ƙarshe, babban canji a ƙirar ASR.

Me yasa DeepSpeech ke amfani da maimaituwar layi na bidirectional?

RNN na bi-directional yana aiwatar da jeri a bangarorin biyu, don haka kowane fitarwa yana fa'ida daga mahallin sauti mai kewaye, yana inganta daidaito.

Wadanne nau'ikan abubuwan shigar da DeepSpeech ke yawan aiki akai?

Samfurin yana cinye fasalulluka na sauti na firam kamar spectrograms ko MMFCs kuma yana fitar da yuwuwar halayen kowane mataki na lokaci.

Abin da ake ƙarawa sau da yawa a lokacin yanke lamba don inganta zaɓin kalmar DeepSpeech da rubutun kalmomi?

Haɗa fitowar sauti tare da ƙirar harshe na waje yayin yanke hukunci yana taimaka wa tsarin samar da ƙarin sahihan kalmomi da haruffa.