Robust Cochlear-Model-Based Speech Recognition

Russo, Mladen; Stella, Maja; Sikora, Marjan; Pekić, Vesna

doi:10.3390/computers8010005

Cited by 16 publications

(17 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Some models are biologically detailed and accurately capture particular response properties of the auditory nerve (6,(11)(12)(13)(14)(15)(16), while others are abstracted approximations of the signal transformation in the auditory periphery (17)(18)(19). Some have been used to provide inputs for models of auditory neurons (17-21), to generate perceptual models (22), and in machine processing of sounds (2,23). However, few attempts (24) have been made to determine which cochlear models best describe the computational impact of the auditory periphery on neural responses in mammalian auditory cortex, although more progress has been made in the avian auditory system (25).…”

mentioning

confidence: 99%

Simple transformations capture auditory input to cortex

Rahman

Willmore

King

et al. 2020

Proc. Natl. Acad. Sci. U.S.A.

View full text Add to dashboard Cite

Sounds are processed by the ear and central auditory pathway. These processing steps are biologically complex, and many aspects of the transformation from sound waveforms to cortical response remain unclear. To understand this transformation, we combined models of the auditory periphery with various encoding models to predict auditory cortical responses to natural sounds. The cochlear models ranged from detailed biophysical simulations of the cochlea and auditory nerve to simple spectrogram-like approximations of the information processing in these structures. For three different stimulus sets, we tested the capacity of these models to predict the time course of single-unit neural responses recorded in ferret primary auditory cortex. We found that simple models based on a log-spaced spectrogram with approximately logarithmic compression perform similarly to the best-performing biophysically detailed models of the auditory periphery, and more consistently well over diverse natural and synthetic sounds. Furthermore, we demonstrated that including approximations of the three categories of auditory nerve fiber in these simple models can substantially improve prediction, particularly when combined with a network encoding model. Our findings imply that the properties of the auditory periphery and central pathway may together result in a simpler than expected functional transformation from ear to cortex. Thus, much of the detailed biological complexity seen in the auditory periphery does not appear to be important for understanding the cortical representation of sound.

show abstract

mentioning

confidence: 99%

Simple transformations capture auditory input to cortex

Rahman

Willmore

King

et al. 2020

Proc. Natl. Acad. Sci. U.S.A.

View full text Add to dashboard Cite

show abstract

“…Tapis gammatone dilaporkan memberi perkiraan yang baik dari tapis auditori manusia untuk urutan tapis 3, 4, dan 5 [28]. Berbeda dengan spectrogram, tapis gammatone memiliki bandwidth tidak konstan di seluruh kanal frekuensi, yaitu Equivalent Rectangular Bandwidth (ERB) [29], [30]. Secara umum, ERB dari tapis gammatone direpresentasikan dengan (2).…”

Section: A Fitur Cochleagramunclassified

“…Dalam eksperimen ini, digunakan implementasi tapis gammatone dari Slaney [30], [31], dengan jumlah 64 standar tapis berjarak 50 Hz hingga 8 kHz (sinyal wicara pada 16 kHz).…”

Section: A Fitur Cochleagramunclassified

Kombinasi Fitur Multispektrum Hilbert dan Cochleagram untuk Identifikasi Emosi Wicara

Gumelar

Yuniarno

Anggraeni

et al. 2020

JNTETI

View full text Add to dashboard Cite

Dalam interaksi perilaku sosial, suara manusia menjadi salah satu saluran utama pembawa atribut ekspresi emosi kondisi mentalnya. Suara manusia merupakan hasil olah vokal yang tersusun dengan disertai urutan kata demi kata, hingga menghasilkan kalimat dalam rupa pola wicara yang memiliki makna ekspresi kondisi psikologisnya. Pola tersebut memberikan karakteristik khusus untuk proses identifikasi biometrik yang menggunakan pola wicara. Teknik visualisasi berupa citra spektrum telah terbukti mampu memberikan representasi hasil olah sinyal wicara. Makalah ini mengidentifikasi jenis emosi pada wicara menggunakan kombinasi fitur multi spektrum Hilbert dan cochleagram. Spektrum Hilbert merepresentasikan hasil transformasi Hilbert-Huang (HHT) untuk memproses sinyal emosi wicara yang nonlinear dan nonstasioner secara instan dengan fungsi mode intrinsik. Dengan meniru cara kerja komponen telinga luar dan tengah, sinyal emosi wicara dipecah menjadi frekuensi yang berbeda secara alami dengan hasil representasinya berupa cochleagram. Kedua masukan berupa spektrum wicara diproses menggunakan metode Convolutional Neural Networks (CNN) yang dikenal terbaik dalam mengenali data citra karena merepresentasikan mekanisme kerja retina manusia, serta metode Long Short-Term Memory (LSTM). Berdasarkan hasil uji coba dengan tiga himpunan data (dataset) publik emosi wicara yang terbagi ke dalam delapan kelas emosi, diperoleh akurasi sebesar 90,97% dengan CNN dan 80,62% dengan LSTM.

show abstract

“…Similarly, the digital filterbank resembles the processing of basilar membrane in auditory modeling. Each bandpass filter is simulated the frequency characteristics of the basilar membrane [14]. The human hearing is the most sensitive to frequencies between 2000-5000Hz and less sensitive in high frequency region.…”

Section: B Gammatone Filterbank Analysismentioning

confidence: 99%

A Front-End Technique for Automatic Noisy Speech Recognition

Naing

Hidayat

Hartanto

et al. 2020

2020 23rd Conference of the Oriental COCOSDA International Committee for the Co-Ordination and Standardisation of Speech Databa

View full text Add to dashboard Cite

The sounds in a real environment not often take place in isolation because sounds are building complex and usually happen concurrently. Auditory masking relates to the perceptual interaction between sound components. This paper proposes modeling the effect of simultaneous masking into the Mel frequency cepstral coefficient (MFCC) and effectively improve the performance of the resulting system. Moreover, the Gammatone frequency integration is presented to warp the energy spectrum which can provide gradually decaying the weights and compensate for the loss of spectral correlation. Experiments are carried out on the Aurora-2 database, and frame-level cross entropy-based deep neural network (DNN-HMM) training is used to build an acoustic model. While given models trained on multi-condition speech data, the accuracy of our proposed feature extraction method achieves up to 98.14% in case of 10dB, 94.40% in 5dB, 81.67% in 0dB and 51.5% in -5dB, respectively.

show abstract

Robust Cochlear-Model-Based Speech Recognition

Cited by 16 publications

References 39 publications

Simple transformations capture auditory input to cortex

Simple transformations capture auditory input to cortex

Kombinasi Fitur Multispektrum Hilbert dan Cochleagram untuk Identifikasi Emosi Wicara

A Front-End Technique for Automatic Noisy Speech Recognition

Contact Info

Product

Resources

About