Txhim kho Feature Parameter Extraction Los ntawm Kev Hais Lus Qhia Siv Tshuab Kawm Algorithm (2)
May 30, 2023
3.7. Dataset Dataset sau thiab tiam txheej txheem tau ua raws li hauv qab no. Hauv txoj kev tshawb no, cov ntaub ntawv nrog 120 h ntawm lub suab tau siv rau kev cob qhia qauv. Cov ntaub ntawv suav nrog cov lus kaw suab uas muaj cov kab lus uas muaj ntau tshaj 15 lo lus nrog tag nrho ntev ntawm kwv yees li 120 h.
Suab puam ginseng
Tsis tas li ntawd, cov ntaub ntawv suav nrog ntau cov ntawv sib txawv rau siv los tsim cov qauv lus. Tshaj 90,650 cov lus hais, 415,780 lo lus, thiab 65,810 cov lus tshwj xeeb uas tau suav nrog hauv cov ntawv sau tau raug sau, ua rau nyob ib puag ncig 120 h ntawm cov ntaub ntawv hais lus. Peb faib cov ntaub ntawv rau hauv kev cob qhia, kev lees paub, thiab cov ntawv xeem. Cov ntaub ntawv txheeb cais
tau qhia hauv Table 3.
Table 3. Cov ntaub ntawv specifications.
Peb faib cov ntaub ntawv mus rau hauv peb lub folders sib raug rau cov kev cob qhia, kev siv tau, thiab cov ntawv xeem. Txhua lub nplaub tshev muaj cov kaw suab thiab cov ntawv sau tseg. Cov ntaub ntawv suab thiab cov ntaub ntawv sib txuas lus yog tib yam, tshwj tsis yog tias cov ntaub ntawv kaw tseg tau khaws cia ua WAV cov ntaub ntawv, thaum cov ntaub ntawv sau cia yog khaws cia ua TXT cov ntaub ntawv siv UTF-8 encoding. Tag nrho cov ntawv sau yog sawv cev siv cov tsiaj ntawv Latin uas muaj 29 cov ntawv thiab cov cim apostrophe. Txhawm rau tiv thaiv overfitting, peb siv cov ntaub ntawv augmentation cov tswv yim raws li kev tshaj tawm perturbation thiab spectral li.
Cistanche deserticola
4. Txoj Kev Npaj
Ib txoj haujlwm tseem ceeb rau cov neeg ua haujlwm thaum tsim cov kev paub txog kev hais lus yog kev tsim cov qauv zoo tshaj plaws rau cov lus qhia parametric ntawm cov lus qhia [45]. Txoj kev no ua rau muaj kev sib cais zoo ntawm cov suab thiab cov lus hais thaum ua kom cov neeg hais lus tsis hnov tsw rau cov qauv kev hais lus thiab kev hloov pauv hauv ib puag ncig acoustic. Feem ntau qhov yuam kev hauv kev paub lo lus yog tshwm sim los ntawm kev hloov pauv ntawm lub suab ntawm lub teeb liab vim muaj kev hloov pauv hauv lub microphone lossis qhov sib txawv ntawm lub suab ntawm cov lus [46]. Lwm qhov ua rau yuam kev yog random nonlinear deformations ntawm cov duab spectrum, uas ib txwm muaj nyob rau hauv cov lus qhia ntawm tus neeg hais lus [47,48]. Yog li ntawd, ib qho ntawm cov haujlwm tseem ceeb tshaj plaws hauv kev tsim cov kev paub txog kev hais lus zoo yog kev xaiv cov neeg sawv cev uas txaus rau cov ntsiab lus ntawm cov teeb liab txheeb xyuas nrog rau qhov tsis hnov tsw ntawm lub suab ntawm cov neeg hais lus thiab ntau qhov chaw acoustic.
Cistanche ntxiv nyob ze kuv-Kev Txhim Kho Kev Nco
Lub kaw lus uas yog siv rau extracting feature parameters feem ntau muaj cov hauv qab no yuav tsum tau. Cov ntsiab lus ntawm cov ntaub ntawv, uas yog, cov txheej txheem ntawm cov yam ntxwv, yuav tsum ua kom ntseeg tau tias kev txheeb xyuas qhov tseeb ntawm cov lus hais tau paub. Tsis tas li ntawd, lub suab nrov, uas yog, qhov siab tshaj plaws compression ntawm lub suab teeb liab, thiab qhov tsis sib xws ntawm cov kev ntsuas yuav tsum raug txo kom tsawg. Kev ywj pheej los ntawm tus neeg hais lus yuav tsum tau ua tiav, uas yog, qhov kev tshem tawm ntau tshaj plaws ntawm cov ntaub ntawv hais txog cov yam ntxwv ntawm tus neeg hais lus los ntawm vector ntawm cov cim. Thaum kawg, homogeneity, uas yog hais txog qhov tsis muaj qhov sib txawv nruab nrab ntawm qhov sib txawv thiab muaj peev xwm siv cov metrics yooj yim los txiav txim siab qhov sib txawv ntawm cov cim, yuav tsum tau muab [49]. Txawm li cas los xij, nws tsis yog ib txwm ua tau kom ua tiav txhua qhov kev xav tau ib txhij vim tias cov kev xav tau sib txawv. Cov lus piav qhia parametric ntawm cov ntsiab lus hais lus yuav tsum muaj cov ncauj lus kom ntxaws kom paub qhov txawv lawv ntseeg tau thiab yuav tsum yog laconic li sai tau.
Superman tshuaj ntsuab cistanche
Hauv kev xyaum, cov teeb liab hais lus uas tau txais los ntawm lub microphone yog digitized ntawm tus qauv ntawm 8 txog 22 kHz. Cov naj npawb ntawm qhov tseem ceeb tau muab faib ua cov lus hais (thav duab) nrog lub sijhawm ntawm 10 mus rau 30 ms, uas sib haum rau qhov chaw hais lus quasi-stationary. Ib tug vector ntawm nta yog xam los ntawm txhua tus ncej, uas yog tom qab siv nyob rau hauv lub acoustic theem ntawm kev paub hais lus. Tam sim no, ntau txoj hauv kev muaj nyob rau hauv parametric sawv cev ntawm cov cim raws li kev txheeb xyuas autocorrelation, kho vajtse linear filtering, spectral tsom xam, thiab LPC. Txoj hauv kev zoo tshaj plaws rau kev hais lus parameterization yog kev tsom xam spectral ntawm cov teeb liab tawg thiab kev suav ntawm lawv cov cepstral coefficients.
MFCCs tau siv los ua cov lus qhia rau cov teeb liab hais lus [41]. Cov yam ntxwv no tau siv dav hauv kev paub hais lus thiab ua raws li ob lub ntsiab lus tseem ceeb: cepstral thiab Mel nplai. Lub ntsiab zoo ntawm cov algorithm yog nws cov theem siab ntawm kev paub thiab yooj yim ntawm kev hais lus. Cov yam ntxwv MFCC raug cais tawm ntawm cov cim hais lus kaw. MFCC algorithm siv cov txiaj ntsig ntawm phonogram thiab spectrum hloov algorithms. Cov algorithm classical uas yog siv los xam cov MFCCs yog piav nyob rau hauv daim duab 5.
Daim duab 5. Classical tswvyim rau xam MFCCs.
Txoj kev tshawb no nthuav tawm txoj hauv kev sai rau kev rho tawm cov haujlwm tsis ua haujlwm los ntawm cov teeb liab hais lus. Lub tswv yim algorithm rau kev suav ceev ntawm MFCCs yog qhia hauv daim duab 6.
Daim duab 6. Cov txheej txheem rau kev suav MFCCs.
Peb xav txog qhov ua tiav ib ntus ntawm qhov kev thov algorithm rau kev nrawm nrawm ntawm cov haujlwm tsis ua haujlwm los ntawm cov teeb liab hais lus
4.1. Faib rau hauv Frames
Tom qab kev lim dej ua ntej, cov teeb liab hais lus tau muab faib ua 16 ms ntas. Txhua tus ncej (tshwj tsis yog thawj zaug) muaj qhov kawg 10 ms ntawm tus ncej dhau los. Cov txheej txheem no txuas ntxiv mus txog thaum kawg ntawm lub teeb liab. Hauv qhov kev tshawb fawb no, vim tias qhov piv txwv ntawm cov teeb liab hais lus yog 16 kHz, tus ncej ntev yog N=256, thiab qhov ntev offset yog M=160. Qhov sib tshooj yog 62.5 feem pua ntawm cov ncej ntev. Kev pab them nqi ntawm 50 feem pua rau 75 feem pua ntawm cov ncej ntev yog feem ntau pom zoo.
4.2. Hanning Qhov rai thiab txo tus nqi
Hanning qhov rais loj ntawm 1D tau siv. Lub qhov rais Hanning tseem hu ua lub qhov rais tsa cosine. Lub qhov rais Hanning tuaj yeem xav tias yog qhov sib npaug ntawm cov zaus spectrum ntawm peb lub sijhawm lub qhov rais. Nws tuaj yeem siv sab lobes kom tshem tawm ib leeg tawm, tshem tawm kev cuam tshuam ntau zaus thiab lub zog to. Hanning qhov rais yog qhov muaj txiaj ntsig zoo ntawm qhov rais.
Cistanche ntxiv nyob ze kuv-Kev Txhim Kho Kev Nco
Nyem qhov no mus saib Cistanche Txhim Kho Kev Nco thiab Tiv Thaiv Alzheimer's Disease cov khoom
【Nug ntxiv】 Email: cindy.xue@wecistanche.com / Whats App: 0086 18599088692 / Wechat: 18599088692
Lub thawv hnyav yog siv los txo qhov kev sib txawv thiab du ntawm tus kheej thav ntawv. Lub teeb liab ntab uas tau tshuaj xyuas hauv qhov kev tshawb fawb no muaj lub suab nrov. Qhov loj ntawm lub suab nrov yog txiav txim siab los ntawm qhov loj ntawm lub suab ntshiab ntawm zaus f, uas yog lim los ntawm Hanning qhov rais.
Ib qho tseem ceeb ntawm lub qhov rais no yog tias nws teev ciam teb ntawm cov thav duab rau xoom. Nyob rau hauv rooj plaub no, lub zog luv luv tuaj yeem suav nrog thaum hla lub qhov rais, thiab lawv tuaj yeem hloov mus los ntawm qhov sib lawv liag uas yog siv los xam qhov qis tshaj amplitude energies. Lub hom phiaj yog tshem tawm cov hluav taws xob qis ntawm lub teeb liab los ntawm kev suav cov teeb liab lub zog thaum smoothing lub teeb liab ntawm lub qhov rais no. Cov txheej txheem no yuav tsum muaj cov teeb liab hluav taws xob sib npaug hauv qab no:
qhov twg En yog lub zog ntawm cov teeb liab input, thiab xi yog tus nqi teeb liab.
Ntxiv rau qhov txheej txheem qhov rais siv (11), lub teeb liab tau ua tiav nyob rau hauv cov kauj ruam tom ntej, uas ua rau txo qis tus naj npawb ntawm qhov tseem ceeb uas nkag mus rau lub processor. Daim duab 7 nthuav qhia cov kev sib txuas ua ke algorithm.
Qhov loj ntawm lub qhov rais sawv cev rau ntau qhov qauv thiab lub sijhawm. Nws yog lub ntsiab parameter ntawm kev tsom xam. Qhov loj ntawm lub qhov rais nyob ntawm qhov tseem ceeb zaus, kev siv zog thiab kev hloov hauv lub teeb liab.
Daim duab 7. Hanning qhov rais algorithm rau tshem tawm qhov ntsiag to. 4.3. Short-Time Fourier Transform (STFT) Hloov Ib qho kev nkag siab zoo ntawm lub ntsiab lus ntawm qhov siab lossis qhov siab qis. STFT yog Fourierrelated hloov pauv uas yog siv los txiav txim siab qhov sinusoidal zaus thiab theem cov ntsiab lus ntawm cov ntu hauv zos ntawm lub teeb liab thaum nws hloov pauv lub sijhawm. Hauv kev xyaum, STFT kev suav suav suav nrog kev faib lub sijhawm ntev dua rau hauv ntu luv ntawm qhov sib npaug ntawm qhov ntev, ollowed los ntawm kev suav cais ntawm Fourier hloov ntawm txhua ntu luv.Qhov no qhia txog Fourier spectrum ntawm txhua ntu luv.Discrete-time signals yog siv hauv kev xyaum. Qhov sib thooj lub sij hawm-frequency conver-sion yog ib tug discrete Fourier hloov dua siab tshiab, uas piav txog qhov ntev ntawm lub teeb liab Xn raws li tus neeg sawv cev ntawm complex nqi zaus sau ntawm N coefficients. STFT, uas piav qhia txog kev hloov pauv ntawm cov khoom siv zaus dhau sijhawm, yog ib qho ntawm cov cuab yeej siv dav tshaj plaws rau kev hais lus thiab kev ua haujlwm [50]. Zoo ib yam li cov spectrum nws tus kheej, ib qho txiaj ntsig ntawm STFT yog tias nws qhov tsis muaj kev txhais lub cev thiab kev xav. STFT feem ntau pom tau siv lub log spectra 20log10 (X(h, k)). Xws li 2D log spectra tuaj yeem raug saib siv daim ntawv qhia thermal hu ua spectrogram. Nyob rau theem thib peb ntawm algorithm, STFT spectral switching txheej txheem yog siv rau cov thav ntawv uas dhau los ntawm qhov hnyav qhov rais. STFT ntawm lub teeb liab tau txais los ntawm kev qhib lub qhov rais thiab txiav txim siab DFT ntawm txhua lub qhov rais. Tshwj xeeb, kev hloov pauv rau Xn thiab Wn qhov rais ntawm lub teeb liab tawm tswv yim yog txiav txim siab raws li hauv qab no:
qhov twg lub k-index sib haum rau qhov zaus qhov tseem ceeb, thiab wn yog lub qhov rais ua haujlwm, uas feem ntau yog Hanning qhov rais lossis Gaussian qhov rais uas nyob nruab nrab ntawm xoom.
4.4. Mel Hloov
Nyob rau theem plaub, lub teeb liab uas hloov mus rau qhov zaus band yog muab faib ua ntau yam uas siv cov duab peb sab. Lub fifilter ciam teb yog suav nrog cov chalk zaus. Kev hloov mus rau chalk zaus fifield yog raws li hauv qab no:
qhov twg f yog qhov ntau zaus.
Qhov rov qab hloov pauv tau txiav txim raws li hauv qab no:
Xav txog NN raws li tus naj npawb ntawm cov fifilters (26 ffilters feem ntau siv) thiab flflows, siab npaum li qhov ntau zaus hauv kev kawm. Qhov ntau yam no tau pauv mus rau Mel nplai thiab muab faib ua NN sib npaug sib npaug sib npaug. Lub linear zaus-tsim ciam teb yog txiav txim nyob rau hauv lub fifield, whereas lub luj coeffificients uas tau raws li fifiltration yog denoted los ntawm H. Tom qab, cov fifilters yog siv rau lub square modulus ntawm cov coefficients tau los ntawm Fourier transform. Cov txiaj ntsig uas tau txais yog logarithmic vim yog cov lus hauv qab no:
Tus lej tus nqi decomposition algorithm yog siv nyob rau theem kawg ntawm kev suav cov MFCCs.
5. Cov txiaj ntsig kev sim
Peb tau siv thiab sim cov txheej txheem uas tau hais tseg hauv Visual Studio 2019 C ntxiv rau ntawm PC nrog 4.90 GHz CPU, 32 GB ntawm RAM, thiab ob Nvidia GeForce 2080Ti GPUs, raws li qhia hauv Table 4. Lub kaw lus tau sim hauv ntau qhov chaw sib txawv rau ntsuas qhov ua tau zoo ntawm cov teeb liab feature extraction txoj kev. Hauv kev sim, thaum lub sijhawm ua haujlwm ntawm lub algorithm (Daim duab 8), thaum qhov sib lawv liag ntawm qhov chaw ntawm lub teeb liab yog n=15, tus naj npawb ntawm cov nqi raug txo los ntawm 40-50 feem pua, thiab lub sijhawm ua haujlwm tau nce 1. 2-xov. Yog li ntawd, qhov algorithm no tau nthuav tawm qhov ua tau zoo dua qub. Ntxiv mus, qhov algorithm no tau ua kom muaj kev sib cais thiab tshem tawm ntawm thaj chaw ntawm kev ntsiag to thaum hla lub qhov rais Hanning.
Muaj ob peb txoj hauv kev uas muaj peev xwm zam kom tsis txhob pov tseg ntawm lub cim xeeb bandwidth. Peb tawm tswv yim kev daws teeb meem tshiab uas ua rau kom kev ua tau zoo ntawm kev suav los ntawm kev sib piv qhov loj ntawm cov teeb liab thav duab mus rau qhov thaiv qhov loj ntawm cache nco. Hom kev ua kom zoo no tuaj yeem cuam tshuam rau tag nrho kev ua haujlwm sib luag. Txawm li cas los xij, nws tuaj yeem siv rau hauv kev ua cov teeb liab digital los ntawm kev faib cov teeb liab rau hauv thav ntawv los ntawm kev siv ntawm multicore processors. Txawm li cas los xij, hauv kev xyaum, kev xaiv feem ntau yog ua rau me me, sib haum rau qhov dav ntawm cov ntaub ntawv tsheb npav uas txuas lub cim xeeb cache rau lub cim xeeb loj thiab qhov loj ntawm nws qhov thaiv. Peb txoj kev siv qhov zoo tshaj plaws ntawm kev siv cov kev nco no hauv kev suav sib txuas. Lub koom haum ntawm lub cim xeeb cache plays lub luag haujlwm tseem ceeb hauv kev ua haujlwm sib luag thaum sib faib cov ntaub ntawv rau hauv cov kwj deg. Tshwj xeeb, muaj cov teebmeem vector-matrix hauv kev ua cov teeb liab digital thiab qhov loj ntawm lawv cov kwj yuav tsum raug kho raws li qhov loj ntawm cov cache blocks. Qhov no tuaj yeem ua tiav siv txoj kev npaj, raws li tau piav qhia hauv daim duab 9.
Table 4. Cov ncauj lus kom ntxaws ntawm qhov kev sim teeb tsa.
daim duab 8. (a) Pib pib lub teeb liab thiab (b) zoo li lub teeb liab tom qab thov cov txheej txheem thov.
Daim duab 9. Parallel xam qauv siv RK3288 processors.
Hauv seem no, peb tham txog kev txheeb xyuas ntau los sib piv qhov kev ua tau zoo ntawm cov tshuab ferent. Peb piv peb txoj kev nrog cov paub hais lus paub zoo algo rithms raws li kev kawm tob. Kev ntsuas ntsuas yog qhov tseem ceeb rau kev sib txuas lus ntau lub tswv yim rau kev paub txog kev hais lus thiab ntsuas qhov ua tau zoo ntawm ntau txoj hauv kev. Txawm hais tias peb siv cov txiaj ntsig ntawm lwm cov kev tshawb fawb rau kev sib piv, peb tsis paub tseeb tias lawv puas muaj tseeb vim tias cov lej cim thiab cov ntaub ntawv ntawm cov txheej txheem no tsis muaj rau pej xeem los txheeb xyuas qhov ua tau zoo. Daim duab 10 qhia txog qhov tshwm sim ntawm qhov rov txav ntawm qhov chaw ntsiag to thaum lub sijhawm dhau los ntawm kev hais lus teeb liab tawg los ntawm Han ning qhov rais raws li lub tswv yim ceev algorithm. Cov txiaj ntsig ceev tau los ntawm kev txheeb xyuas tau nthuav tawm hauv Table 5.
Daim duab 10. k tus nqi ntawm KNN algorithm (nrog kev xaiv tshwj xeeb).
Table 5. Kev sim cov txiaj ntsig ntawm txoj kev npaj.
Ib txoj hauv kev txiav txim siab tau txiav txim siab ua kom muaj txiaj ntsig ntawm cov neeg nyob ze uas muab cov txiaj ntsig zoo tshaj plaws hauv lub hauv caug. Qhov kev qhia tshwj xeeb npog 1-25. Nyob rau hauv daim duab 10, daim duab ntawm KNN algorithm nrog kev xaiv feature tau siv. Thaum cov duab tau tshuaj xyuas, thaum tus nqi hauv zej zog yog 1 thaum pib, qhov kev cob qhia raug ntau dua li qhov kev xeem raug. Hauv KNN algorithm tsim los siv cov yam ntxwv xaiv los ntawm kev sib raug zoo, qhov tseeb ntawm tus qauv tau txiav txim siab yog 99.15 feem pua hauv cov ntaub ntawv qhia kev cob qhia thiab 97.35 feem pua hauv cov ntaub ntawv xeem.
Lo lus yuam kev tus nqi (WER) lossis tus cwj pwm yuam kev feem ntau yog siv los ntsuas qhov tseeb ntawm kev rho tawm ntawm cov teeb liab hais lus. Cov no yog cov hom phiaj matrices uas muaj txiaj ntsig zoo rau kev sib piv ncaj ncees ntawm cov txheej txheem kev lees paub. Hauv peb cov kev tshawb fawb yav dhau los [51–56], peb suav cov ntsuas xws li F-measure (FM), precision, thiab rov qab. FM yog qhov hnyav nruab nrab uas ntsuas qhov ntsuas nruab nrab ntawm qhov tseeb thiab rov qab cov nqi. Qhov tseeb yog qhov piv ntawm tus naj npawb ntawm qhov raug kwv yees qhov kev soj ntsuam zoo rau tag nrho cov naj npawb ntawm cov kev soj ntsuam zoo. Kev rov qab los yog qhov piv ntawm cov lej ntawm qhov raug kwv yees qhov kev soj ntsuam zoo rau tag nrho cov kev soj ntsuam hauv chav kawm tiag tiag, raws li qhia hauv (9). Cov kab zauv hauv qab no tuaj yeem siv los xam qhov nruab nrab precision thiab rov qab cov nqi ntawm cov txheej txheem rho tawm feature:
qhov twg TP qhia txog tus naj npawb ntawm qhov zoo, FP qhia tus lej ntawm qhov tsis zoo, thiab FN qhia txog tus lej ntawm qhov tsis zoo.
FM suav nrog (10), txiav txim siab qhov tseeb thiab rov qab.
Qhov nruab nrab FM, rov qab, thiab qhov tseeb ntawm txoj kev npaj yog 98.4 feem pua. Kev kuaj pom tsis tseeb tshwm sim hauv 1.6 feem pua ntawm cov neeg mob vim tsis xav tau lub suab nrov ntawm cov cim ntawm lub microphone. Qhov ntau ntawm tus qauv qhov tseeb yog nyob nruab nrab ntawm 0 thiab 1, thiab cov qhab nia ntsuas ntsuas tau mus txog lawv qhov zoo tshaj plaws ntawm 1. Kev ntsuam xyuas ntawm peb txoj kev thiab lwm yam kev tshaj tawm tsis ntev los no hais txog kev rho tawm cov ntsiab lus tau nthuav tawm hauv Table 6. Tib tus lej ntawm cov yam ntxwv tau siv rau kev sib piv ncaj ncees. Tag nrho ntawm 325 tus qauv hais lus los ntawm txhua pab pawg tau txheeb xyuas los ntawm cov ncauj lus uas muaj keeb kwm zoo sib xws. Cov teebmeem ntawm qhov sib txawv ntawm qhov ntev raws li tus naj npawb ntawm cov tsev txhab nyiaj fifilter hauv MFCC thiab cov ncej sib txawv nyob rau hauv qhov kev txiav txim ntawm LPC kuj tau kuaj xyuas kom raug.
Table 6. Qhov tseeb qhov tseeb ntawm kev hais lus feature extraction.
Raws li tau hais dhau los, WER yog qhov kev ntsuas tshaj plaws rau kev paub txog kev hais lus. Nws yog xam los ntawm kev sib piv cov ntaub ntawv sau tseg nrog cov zis ntawm tus neeg hais lus. Raws li qhov kev sib piv no, nws tuaj yeem suav cov lej ntawm qhov yuam kev, uas feem ntau yog peb pawg: (1) kev ntxig thaum ib lo lus tsis muaj nyob rau hauv kev siv nyob rau hauv cov zis ntawm kev paub hais lus tsis siv neeg (ASR), (2) tshem tawm thaum ib lo lus ploj lawm hauv ASR cov zis, thiab (3) hloov thaum ib lo lus tsis meej pem nrog lwm lo lus. WER tuaj yeem suav tau raws li hauv qab no.
qhov twg S yog tus naj npawb ntawm kev hloov ntawm cov lus uas tsis raug lees paub, D yog tus lej ntawm kev tshem tawm, Kuv yog tus lej ntawm cov lus ntxig, thiab N yog tus lej ntawm cov lus hauv cov ntaub ntawv siv. Qhov teeb meem tseem ceeb hauv kev suav cov qhab nia no yog kev sib tw ntawm ob lo lus sib lawv liag. Qhov no tuaj yeem txiav txim siab los ntawm kev ua haujlwm dynamic siv Levenshtein nrug [67].
Raws li Table 6, peb tau ua ib qho kev txheeb xyuas los qhia qhov tseeb ntawm qhov nruab nrab ntawm cov kev sib piv uas siv WER kev ntsuas ntsuas ntsuas, raws li tau piav qhia hauv daim duab 11. Qhov kev txhim kho feature extractor tau txais qhov tseeb ntawm kwv yees li 98.4 feem pua, whereas lwm txoj hauv kev tau txais qhov tseeb ntawm 78% thiab 96%. Peb siv cov txiaj ntsig tau muab rau hauv cov ntaub ntawv cuam tshuam rau kev sib piv; Txawm li cas los xij, qhov tseeb ntawm cov txiaj ntsig no tsis yooj yim txheeb xyuas tau vim tias cov lej cim thiab cov ntaub ntawv ntawm cov txheej txheem no tsis muaj rau pej xeem kom paub meej tias lawv qhov kev ua tau zoo tiag tiag. Txawm li cas los xij, nyob rau hauv rooj plaub ntawm cov qauv txheej txheem, txoj kev npaj tau sim ua kom pom qhov zoo tshaj plaws ntawm kev hais lus zoo tshaj plaws los ntawm kev txo cov sij hawm suav, txawm tias cov ntaub ntawv hais lus nrov nrov lossis tsis zoo.
Daim duab 11. Cov txiaj ntsig ntawm kev hais lus teeb liab feature rho tawm txoj hauv kev uas siv cov duab ntsug.
Ntxiv mus, peb tau soj ntsuam cov txiaj ntsig tsis tseeb-zoo ntawm cov kev xaiv. Raws li tuaj yeem pom los ntawm daim duab 12, qhov kev thov mus kom ze muaj qhov yuam kev tsawg tshaj plaws. Tsis tas li ntawd, qhov ua tau zoo heev thaum sib piv txoj kev suav qhov tseem ceeb txo qis lub suab teeb liab feature xaiv thiab rho tawm yuam kev. Overfiftitting yog ib qho ntawm cov teeb meem tseem ceeb thaum lub sij hawm kev cob qhia, thiab yuav luag txhua tus qauv kev kawm tshuab raug kev txom nyem los ntawm nws. Peb tau sim txo qis kev pheej hmoo overfiftitting siv cov txheej txheem xaiv feature uas lub hom phiaj hloov pauv qhov tseem ceeb ntawm cov yam ntxwv uas twb muaj lawm hauv cov ntaub ntawv thiab muab pov tseg qhov tseem ceeb (tsis muaj cov yam ntxwv tshiab tsim).
Daim duab 12. Pom tau cov txiaj ntsig ntawm kev hais lus tsis tseeb-zoo teeb liab feature extraction thwmsim.
Table 7 qhia txog kev ua tau zoo ntawm cov txheej txheem uas tau siv nyob rau hauv kev paub txog kev hais lus raws li cov khoom sib txawv. Peb txoj kev npaj tsis raug kev txom nyem los ntawm qhov tsis xav tau thiab tsis tsim nyog lub suab nrov tom qab thiab tsis cuam tshuam los ntawm cov neeg tsis zoo ntawm cov suab nrov xws li cov suab nrov nrov, cov suab nrov uas ua rau mob caj pas, lossis txawm tias lub suab los ntawm tib neeg nrog lub suab tag. Peb txoj kev tsom mus kov yeej qhov teeb meem ntawm cov cuab yeej kaw tsis txaus, suab nrov tom qab, suab nrov thiab cov lus hais, thiab ntau lub suab hauv lub suab. Nyob rau hauv ib puag ncig ib puag ncig, cov txiaj ntsig zoo tshaj plaws rau kev kuaj xyuas kom raug thiab tshem tawm cov teeb meem kev hais lus tau txais los ntawm kev siv txoj kev npaj nrog lub sijhawm ua haujlwm txo.
Table 7. Kev tshuaj xyuas ntawm kev hais lus feature nrhiav kom tau thiab tshem tawm kev ua tau zoo siv ntau yam nta.
Cov txiaj ntsig ntawm txoj kev paub txog kev hais lus tau raug cais raws li muaj zog, ib txwm muaj, lossis tsis muaj zog rau xya pawg. Cov txheej txheem muaj zog qhia tau hais tias lub algorithm tuaj yeem kov yeej txhua yam kev sib tw. Hauv qhov sib piv, qhov kev ntsuas ib txwm qhia tau hais tias lub algorithm yuav ua tsis tiav hauv qee zaus vim tias lo lus ciam teb tsis tau txhais ua ntej. Thaum kawg, qhov kev ntsuas tsis muaj zog qhia tau hais tias cov algorithm tsis tuaj yeem ntseeg tau hauv qab lub suab nrov lossis kev vibrations.
6. Kev txwv
Nws yog ib qho nyuaj rau kev txiav txim siab tias cov txheej txheem uas tau hais los txog niaj hnub no tsis muaj qhov tshwm sim luv luv. Peb txoj kev npaj kuj tseem yuav ua rau muaj qhov yuam kev vim muaj ntau lub suab nrov ib puag ncig. Txhawm rau kov yeej qhov teeb meem no, peb tsom kom txo tus naj npawb ntawm cov yam ntxwv hauv cov ntaub ntawv los ntawm kev tsim cov yam ntxwv tshiab los ntawm cov yam ntxwv uas twb muaj lawm [69]. Txij li thaum overfiftitting yog ib qho ntawm cov teeb meem tseem ceeb rau kev cob qhia cov qauv sib txawv thaum lub sij hawm kev sib tw, ntxiv dag zog rau cov ntaub ntawv kev cob qhia los ntawm kev ntxiv cov ntaub ntawv los ntawm cov peev txheej sib txawv tuaj yeem yog ib qho kev daws teeb meem rau kev txhim kho cov txiaj ntsig. Tsis hais txog cov teeb meem uas tau hais los saum no, cov txiaj ntsig kev sim tau qhia tias peb txoj kev muaj zog heev thiab ua haujlwm zoo rau kev hais lus tshwj xeeb tshem tawm cov haujlwm, nrog qhov tseeb nruab nrab ntawm 98.4 feem pua thiab FM ntawm 99.5 feem pua.
7. Cov lus xaus
Ib qho kev ua tau zoo tshiab uas muaj kev ua tau zoo sib luag hauv kev suav nrog kev siv tshuab kev kawm tau raug npaj rau cov tshuab kev paub hais lus. Cov teeb meem nrawm hauv cov tshuab uas muaj cov cuab yeej siv tsawg tsawg tuaj yeem daws tau los ntawm kev faib cov tshuab. Kev suav ceev hauv cov teeb liab lees paub cov tshuab tuaj yeem nce ntxiv, thiab kev ua haujlwm ntawm multicore platforms tuaj yeem txhim kho los ntawm kev tsim thiab siv cov txiaj ntsig zoo thiab ceev algo rithms. Cov txiaj ntsig tau pom tias tus qauv npaj txo qis lub sijhawm ua haujlwm thiab txhim kho cov txiaj ntsig kev rho tawm los ntawm 98.4 feem pua los ntawm kev siv MFCCs zoo. Nws tau raug pom tias cov yam ntxwv uas tsis tshua muaj kev sib raug zoo ntawm cov txiaj ntsig tau muab rho tawm los ntawm kev xaiv feature kuj muaj txiaj ntsig zoo rau kev ua tiav ntawm cov qauv. Kev txheeb xyuas kev txheeb xyuas tau ua tiav ntawm cov ntaub ntawv ua ntej ua ntej, thiab cov ntaub ntawv muaj txiaj ntsig tau tsim los ntawm cov ntaub ntawv siv K-qhov ze tshaj plaws nyob ze (KNN) tshuab kev kawm algorithm.
Cov kev tshawb fawb yav tom ntej yuav tsom mus rau kev txhim kho qhov tseeb ntawm peb txoj hauv kev los ntawm kev siv cov kev kawm sib sib zog nqus thiab txhim kho lub cim xeeb cache ntawm cov txheej txheem multicore los kuaj xyuas thiab rho tawm cov lus qhia tsis muaj qhov tsis zoo. Tsis tas li ntawd, peb npaj los tsim cov qauv kev tshuaj ntsuam xyuas spectral raws li kev ua haujlwm sib luag nrog kev soj ntsuam kev ua tau zoo uas yuav ua rau muaj kev tsim cov khoom siv nrog cov khoom siv qis qis siv Taris hais lus datasets [70] hauv 3D CNN thiab 3D U-Net Ib puag ncig [71– 75]
Cov ntaub ntawv
1. Meng, YJ; Liu, WJ; Zhang, RZ; Du, HS Speech Feature Parameter Extraction thiab Recognition Raws li Interpolation. Appl. Mech. Mater. 2014, 602–605, 2118–2123. [CrossRef] 2. Musaev, M.; Rakhimov, M. Accelerated Training for Convolutional Neural Networks. Hauv Kev Ua Haujlwm ntawm 2020 International Conference on Information Science thiab Communications Technologies (ICISCT), Tashkent, Uzbekistan, 4–6 Kaum Ib Hlis 2020; pp. 1–5. [CrossRef] 3. Ye, F.; Yang, J. A Deep Neural Network Model for Speaker Identification. Appl. Sci. 2021, 11, 3603. [CrossRef] 4. Musaev, M.; Rakhimov, M. Ib txoj hauv kev ntawm Mapping ib qhov thaiv ntawm Lub Cim Tseem Ceeb rau Cache nyob rau hauv Parallel Processing ntawm Cov Lus Qhia. Hauv Kev Ua Haujlwm ntawm 2019 Lub Rooj Sib Tham Thoob Ntiaj Teb ntawm Kev Tshawb Fawb Txog Kev Tshawb Fawb thiab Kev Sib Txuas Lus (ICISCT), Karachi, Pakistan, 9-10 Lub Peb Hlis 2019; p. 1–4. [CrossRef] 5. Jiang, N.; Liu, T. Ib qho kev txhim kho hais lus segmentation thiab pawg algorithm raws li SOM thiab k-lus. lej. Teeb meem. Eng. 2020, 2020, 3608286. [CrossRef] 6. Hu, W.; Yaj, Z.; Chen, C.; Sun, IB; Xie, Q. Ib qho kev vibration segmentation mus kom ze rau ntau qhov kev ua haujlwm ntawm cov lej tswj turret. Cov Txheej Txheem Duab Duab Duab. 2021, 16, 489–496. [CrossRef]
7. Popescu, TD; Aiordachioaie, D. Fault nrhiav tau ntawm dov lub caij bearings siv pom segmentation ntawm vibrating signals. Mech. Syst. Txheej txheem teeb liab. 2019, 116, 370–391. [CrossRef] 8. Shihab, MSH; Aditya, S.; Setu, JH; Imtiaz-Ud-Din, KM; Efat, MIA Hybrid GRU-CNN Feature Extraction Technique for Speaker Identification. Hauv Kev Ua Haujlwm ntawm 2020 23rd International Conference on Computer thiab Information Technology (ICCIT), Dhaka, Bangladesh, 19-21 Kaum Ob Hlis 2020; p. 1–6. [CrossRef] 9. Korkmaz, O.; Atasoy, A. Kev xav paub los ntawm kev hais lus teeb liab siv mel-frequency cepstral coefficients. Hauv Kev Ua Haujlwm ntawm 9th Lub Rooj Sib Tham Thoob Ntiaj Teb ntawm Hluav Taws Xob thiab Hluav Taws Xob Engineering (ELECO), Bursa, Qaib ntxhw, 26-28 Kaum Ib Hlis 2015; Ib., 1254–1257. 10. Ayvaz, UA; Gürüler, H.; Khan, F.; Ahmed, N.; Whangbo, T.; Abdusalomov, A. Tsis siv neeg hais lus paub siv Mel-Frequency Cepstral Coeffificients Los ntawm Tshuab Kawm. CMC-Computer. Mater. Contin. 2022, 71, 5511–5521. 11. Al-Qaderi, M.; Lahamer, E.; Rad, A. A Ob-Level Speaker Identifification System ntawm Fusion ntawm Heterogeneous Classififiers thiab Complementary Feature Cooperation. Sensors 2021, 21, 5097. [CrossRef] 12. Batur Dinler, Ö.; Aydin, N. Ib Qhov Zoo Tshaj Plaws Cov Txheej Txheem Txheej Txheem Raws li Gated Recurrent Unit Recurrent Neural Networks rau Kev Hais Lus Segment Tshawb Nrhiav. Appl. Sci. 2020, 10, 1273. [CrossRef] 13. Kim, H.; Shin, JW Dual-Mic Speech Enhancement Raws li TF-GSC nrog Kev Tiv Thaiv Kev Tiv Thaiv thiab Kev Tiv Thaiv Rov Qab. Appl. Sci. 2021, 11, 2816. [CrossRef] 14. Lee, S.-J.; Kwon, H.-Y. Lub Tswv Yim Ua Ntej Ua Ntej rau Denoising ntawm Kev Hais Lus Cov Ntaub Ntawv Raws li Kev Tshawb Fawb Kev Hais Lus. Appl. Sci. 2020, 10, 7385. [CrossRef] 15. Rusnac, A.-L.; Grigore, O. CNN Architectures thiab Feature Extraction Methods for EEG Imaginary Speech Recognition. Sensors 2022, 22, 4679. [CrossRef] [PubMed] 16. Wafa, R.; Khan, MQ; Malik, F.; Abdusalomov, AB; Cho, YI; Odarchenko, R. Qhov cuam tshuam ntawm Agile Methodology ntawm Kev Ua Haujlwm Zoo, nrog rau Kev Ua Haujlwm Zoo ntawm Tus Neeg Ua Haujlwm Zoo hauv IT Kev Lag Luam ntawm Pakistan. Appl. Sci. 2022, 12, 10698. [CrossRef] 17. Aggarwal, A.; Srivastava, UA; Agarwal, UA; Chaw, N.; Singh, D.; Alnuaim, AA; Alhadlaq, A.; Li, H.N. Ob-txoj kev Feature Extraction rau kev hais lus Emotion paub siv Deep Learning. Sensors 2022, 22, 2378. [CrossRef] [PubMed] 18. Marini, M.; Vanello, N.; Fanucci, L. Optimizing Speaker-Dependent Feature Extraction Parameters los txhim kho kev hais lus tsis siv neeg paub txog kev ua tau zoo rau cov neeg mob dysarthria. Sensors 2021, 21, 6460. [CrossRef] 19. Tiwari, S.; Yaj, A.; Sharma, UA; Almustafa, KM Phonocardiogram Signal Based Multi-Class Cardiac Diagnostic Decision Support System. IEEE Access 2021, 9, 110710–110722. [CrossRef] 20. Mohtaj, S.; Schmitt, V.; Möller, S. A Feature Extraction based Model for Hate Speech Identification. arXiv 2022, arXiv: 2201.04227. 21. Kuldoshbay, A.; Abdusalomov, A.; Mukhiddinov, M.; Baratov, N.; Makhmudov, F.; Cho, YI Ib qho kev txhim kho rau txoj kev tsis siv neeg faib rau cov duab ultrasound siv hauv CNN. Int. J. Wavelets Multiresolution Inf. Txheej txheem. 2022, 20, 2150054. 22. Passricha, V.; Aggarwal, RK Ib tug hybrid ntawm sib sib zog nqus CNN thiab bidirectional LSTM rau tsis siv neeg hais lus paub. J. Intell. Syst. 2020, 29, 1261–1274. [CrossRef] 23. Mukhamadiyev, A.; Khujayarov, I.; Djuraev, O.; Cho, J. Txoj Kev Paub Hais Lus Tsis Siv Neeg Raws Li Txoj Kev Kawm Sib Nrauj rau Uzbek Lus. Sensors 2022, 22, 3683. [CrossRef] [PubMed] 24. Li, F.; Liu, M.; Zhao, Y.; Kong, L.; Dong, L.; Li, X.; Hui, M. Feature extraction thiab classification of heart sound using 1D convolutional neural networks. EURASIP J. Adv. Txheej txheem teeb liab. 2019, 2019, 59. [CrossRef] 25. Chang, L.-C.; Hung, J.-W. Txoj Kev Kawm Ua Ntej ntawm Kev Hais Lus Zoo Tshaj Plaws Raws Li Kev Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Lij Tshaj Tawm acoustic. Appl. Syst. Innov. 2022, 5, 71. [CrossRef] 26. Ramírez, J.; Górriz, JM; Segura, JC Voice Activity Detection. Lub hauv paus ntsiab lus thiab kev lees paub qhov system muaj zog. Hauv Kev Paub Hais Lus Zoo thiab nkag siab; Grimm, M., Kroshel, K., Eds.; I-TECH Kev Kawm thiab Kev Tshaj Tawm: London, UK, 2007; pp. 1–22. 27. Huag, S. DNN Raws li Robust Speech Feature Extraction and Signal Noise Removal Method Siv Kev Txhim Kho Nruab Nrab Prediction LMS Lim rau Kev Paub Hais Lus. J. Converg. Inf. Tshuab. 2021, 11, 1–6. [CrossRef] 28. Abbaschian, BJ; Sierra-Sosa, D.; Elmaghraby, A. Deep Learning Techniques for Speech Emotion Recognition, from Databases to Models. Sensors 2021, 21, 1249. [CrossRef] 29. Rakhimov, M.; Mamadjanov, D.; Mukhiddinov, A. Ib qho Kev Ua Tau Zoo Tshaj Plaws Zoo Tshaj Plaws rau Kev Ua Duab Hauv Kev Sib Tw. Hauv Kev Ua Haujlwm ntawm 2020 IEEE 14th International Conference on Application of Information and Communication Technologies (AICT), Uzbekistan, Tashkent, 7–9 Lub Kaum Hli 2020; pp. 1–5. [CrossRef] 30. Abdusalomov, A.; Mukhiddinov, M.; Djuraev, O.; Khamdamov, UA; Whangbo, TK Automatic Salient Object Extraction Raws li Locally Adaptive Thresholding to Generate Tactile Graphics. Appl. Sci. 2020, 10, 3350. [CrossRef] 31. Abdusalomov, A.; Whangbo, TK Ib qho kev txhim kho rau lub hauv paus kev lees paub txoj kev siv cov txheej txheem tshem tawm duab ntxoov ntxoo rau qhov chaw nyob sab hauv. Int. J. Wavelets Multiresolution Inf. Txheej txheem. 2017, 15, 1750039. [CrossRef] 32. Abdusalomov, A.; Whangbo, TK Kev Tshawb Nrhiav thiab Tshem Tawm Ntawm Cov Duab Ntxoo Ntawm Cov Duab Ntxoo siv cov duab geometry thiab cov ntaub ntawv xim rau cov yeeb yaj kiab sab hauv tsev. Appl. Sci. 2019, 9, 5165. [CrossRef] 33. Mery, D. Computer Vision for X-ray Test; Springer International Publishing: Cham, Switzerland, 2015; p. 271, ISBN 978-3319207469 ib. 34. Mark, S. Speech imagery recalibrates hais lus-xav paub ciam teb. Atten. Kev nkag siab. Psychophys. Xyoo 2016, 78, 1496–1511. [CrossRef] 35. Mudgal, E.; Mukunthaaj, S.; Modak, MU; Rao, YS Template Raws Li Lub Sijhawm Hais Lus Pom Zoo Siv Cov Lim Rau DSP-TMS320F28335. Hauv Kev Ua Haujlwm ntawm 2018 Plaub Lub Rooj Sib Tham Thoob Ntiaj Teb ntawm Kev Sib Txuas Lus Kev Sib Txuas Lus thiab Kev Sib Txuas Lus (ICCUBEA), Pune, Is Nrias teb, 16-18 Lub Yim Hli 2018; p. 1–6. [CrossRef]
