Mel-Frequency Cepstral Coefficients
Coefficients ya Mel-Frequency Cepstral (MFCCs) ni igiteranyo cyimibare igereranya incamake yimiterere yumurongo wijwi nkuko amatwi yabantu abibona.
Incamake
For decades they were the workhorse feature for speech recognition, speaker identification, and music analysis.
Kwibira cyane
MFCCs ihindura igice kigufi cyamajwi mumibare igera kuri 13 ifata timbre yayo. Umuyoboro ufata imiraba, ukayicamo ibice ~ 25ms, ikabara amashanyarazi binyuze muri transformateur ya Fourier, hanyuma ikarwanya umurongo wikurikiranya kugeza ku gipimo cya mel, ikaba ihuza imirongo nkuko cochlea ikora: neza munsi ya 1kHz kandi hejuru cyane. Ingufu za mel zirahuzagurika (kwigana amajwi aranguruye) hanyuma amaherezo ikanyuzwa muri cosine yihariye, ikayitandukanya kandi igahuza amakuru muri coefficient ya mbere. Igisubizo kirakomeye cyane ku rusaku no kuvuga, niyo mpamvu uburyo bwa kera bwa Hidden Markov Model hamwe na Gaussian Mixture Model imvugo yo gushingira kuri MFCC hafi ya hose mbere yo kwiga byimbitse.
Ubushishozi
Igipimo cya mel kigereranya imyumvire hamwe na mel = 2595 log10 (1 + f / 700), bityo intambwe zingana zingana zumvikana neza. Impinduka ya nyuma ya cosine ihinduka (DCT) nintambwe ya 'cepstral': ifata log-mel spektrike nkikimenyetso kandi igatandukanya buhoro buhoro imiterere-yijwi ryinzira (coefficient nkeya ya cepstral, igice dukomeza) hamwe nihuta ryihuta ryihuta (coefficient nyinshi, ubusanzwe zajugunywe), zitandukanya neza indangamuntu ya fonetike mukuvuga.
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 Mel-Frequency Cepstral Coefficients
Imiyoboro yimbitse-iherezo iragenda yiga ibintu biturutse kumurongo wibanze cyangwa log-mel spectrograms, gusimbuka DCT, bityo MFCCs nziza ziragenda zishira muburyo bugezweho bwa ASR. Nyamara baracyakunzwe cyane kuburemere bworoshye, kubikoresho, hamwe namakuru make-yamakuru: ijambo ryibanze ryerekana, ibikorwa byerekana amajwi, gutunga amajwi, hamwe na bioacoustics. Tegereza MFCCs ikomeze nkibikorwa byiza, bisobanurwa nubwo byize imbere-impera yiganje kuri moderi nini.
Gushyira mu bikorwa Isi
Ibiranga Acoustic kubisanzwe HMM-GMM imenyekanisha imvugo nka sisitemu ya mbere ya Sphinx na HTK
Kugenzura abavuga no gutandukana, gutandukanya abavuga kumuhamagaro
Ibyiciro byumuziki gutondeka no gutunga urutoki (Shazam-style ya timbre ihuza)
Kumenya amakosa yimashini cyangwa guhamagarira inyamanswa kumajwi mugukurikirana inganda na bioacoustic
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
Shaka uruhushya rusobanutse rwo gufata amajwi, gukoroniza, no gukoresha.
Ikizamini cyiza mubiganiro bitandukanye hamwe nuburyo bwimbere.
Sobanura igihe umuntu agomba gusuzuma cyangwa kwemeza ibisubizo.
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 Mel-Frequency Cepstral Coefficients 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
Ubuyobozi bukurikira
Mel Spectrograms
Ibibazo bikunze kubazwa
What is Mel-Frequency Cepstral Coefficients?
Coefficients ya Mel-Frequency Cepstral (MFCCs) ni igiteranyo cyimibare igereranya incamake yimiterere yumurongo wijwi nkuko amatwi yabantu abibona. Mu myaka ibarirwa muri za mirongo bari ibikoresho byakazi byo kumenyekanisha imvugo, kumenyekanisha abavuga, no gusesengura umuziki.
'Mel' muri MFCC bivuga iki?
Igipimo cya mel gitsindagira inshuro kuburyo intambwe zingana zumvikana kure cyane kubantu, zashyizwe neza cyane kuri radiyo nto kandi zoroheje cyane.
Ni izihe mpinduka zikoreshwa bwa nyuma kugirango zitange cepstral coefficient?
Ingufu za log-mel zimaze kubarwa, impinduka ya cosine yihariye irayitandukanya kandi igapakira amakuru muri coefficient ya mbere.
Ni ukubera iki MFCCs ikomera cyane mukibuga cyabavuga?
Coefficient ya cepstral yo hasi ifata ibahasha yijwi ryamajwi (ibirimo fonetike) mugihe muremure ifata ikibuga, bityo kugumya hasi ntigishimangira ikibuga.
Hafi ya coefficient za MFCC zingahe zibikwa kumurongo?
Guhitamo bisanzwe ni coefficient zigera kuri 13, rimwe na rimwe zongerwaho na delta zabo, zerekana incamake ya timbre.
Ni ukubera iki igiti gikoreshwa ku mbaraga za mel-band?
Ijwi ryabantu ryumvikana ni logarithmic, so compression ituma ibiranga bihuza neza imyumvire kandi bigahindura urwego rugaragara.