Hearing Is Believing: Biologically Inspired Methods for Robust Automatic Speech Recognition

Stern, Richard M.; Morgan, Nelson

doi:10.1109/msp.2012.2207989

Cited by 50 publications

(22 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, auditory-inspired Constant-Q Cepstral Coefficients (CQCC) also perform better compared to MFCC features specifically in detecting spoof by unit selection speech synthesis system [12], [13]. Such handcrafted features rely on simplified auditory models [14], [15].…”

Section: Introductionmentioning

confidence: 99%

Unsupervised Representation Learning Using Convolutional Restricted Boltzmann Machine for Spoof Speech Detection

2017

View full text Add to dashboard Cite

Speech Synthesis (SS) and Voice Conversion (VC) presents a genuine risk of attacks for Automatic Speaker Verification (ASV) technology. In this paper, we use our recently proposed unsupervised filterbank learning technique using Convolutional Restricted Boltzmann Machine (ConvRBM) as a frontend feature representation. ConvRBM is trained on training subset of ASV spoof 2015 challenge database. Analyzing the filterbank trained on this dataset shows that ConvRBM learned more low-frequency subband filters compared to training on natural speech database such as TIMIT. The spoofing detection experiments were performed using Gaussian Mixture Models (GMM) as a back-end classifier. ConvRBM-based cepstral coefficients (ConvRBM-CC) perform better than hand crafted Mel Frequency Cepstral Coefficients (MFCC). On the evaluation set, ConvRBM-CC features give an absolute reduction of 4.76 % in Equal Error Rate (EER) compared to MFCC features. Specifically, ConvRBM-CC features significantly perform better in both known attacks (1.93 %) and unknown attacks (5.87 %) compared to MFCC features.

show abstract

Section: Introductionmentioning

confidence: 99%

Unsupervised Representation Learning Using Convolutional Restricted Boltzmann Machine for Spoof Speech Detection

2017

View full text Add to dashboard Cite

show abstract

“…Findings about the auditory system have influenced research in automatic speech recognition (ASR), which often resulted in more robust machine listening [1,2]. Although a closer connection between ASR and human speech recognition (HSR) has been promoted earlier to further our understanding of speech processing in humans and machines [3], there is a comparatively small number of studies that bring back ASR technology that profited from auditory insights to better understand HSR; important exceptions are for instance [4] and [5].…”

Section: Introductionmentioning

confidence: 99%

Neural Responses to Speech-Specific Modulations Derived from a Spectro-Temporal Filter Bank

Frye¹,

Micheli

Schepers

et al. 2016

Interspeech 2016

View full text Add to dashboard Cite

This paper analyzes the application of methods developed in automatic speech recognition (ASR) to better understand neural activity measured with electrocorticography (ECoG) during the presentation of speech. ECoG data is collected from temporal cortex in two subjects listening to a matrix sentence test. We investigate the relation of ECoG signals and acoustic speech that has been processed with spectro-temporal filters, which have been shown to produce robust and reliable representations for speech applications. The organization of spectro-temporal filters into a filter bank allows for a straight-forward separation into spectral or temporal only, as well as true spectro-temporal components. We find electrodes positioned over the superior temporal gyrus that is associated with the auditory cortex to show significant specific high gamma activity to fine temporal and spectro-temporal patterns present in speech. This indicates that representations developed in machine listening are a suitable tool for the analysis of biosignals.

show abstract

“…On the other hand, psychoacoustic experiments also demonstrate that the joint spectro-temporal modulations are highly related to speech intelligibility [27] and speech comprehension [11]. The concept of using spectro-temporal modulations has since been adopted in many applications, such as speech intelligibility assessment [28], musical instrument identification [29], and robust feature extraction for automatic speech [30], [31] and speaker recognition [32].…”

Section: Introductionmentioning

confidence: 99%

Robust Voice Activity Detection Algorithm Based on Feature of Frequency Modulation of Harmonics and Its DSP Implementation

Hsu

Cheong

Chi

et al. 2015

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARY This paper proposes a voice activity detection (VAD) algorithm based on an energy related feature of the frequency modulation of harmonics. A multi-resolution spectro-temporal analysis framework, which was developed to extract texture features of the audio signal from its Fourier spectrogram, is used to extract frequency modulation features of the speech signal. The proposed algorithm labels the voice active segments of the speech signal by comparing the energy related feature of the frequency modulation of harmonics with a threshold. Then, the proposed VAD is implemented on one of Texas Instruments (TI) digital signal processor (DSP) platforms for real-time operation. Simulations conducted on the DSP platform demonstrate the proposed VAD performs significantly better than three standard VADs, ITU-T G.729B, ETSI AMR1 and AMR2, in non-stationary noise in terms of the receiver operating characteristic (ROC) curves and the recognition rates from a practical distributed speech recognition (DSR) system.

show abstract

Hearing Is Believing: Biologically Inspired Methods for Robust Automatic Speech Recognition

Cited by 50 publications

References 44 publications

Unsupervised Representation Learning Using Convolutional Restricted Boltzmann Machine for Spoof Speech Detection

Unsupervised Representation Learning Using Convolutional Restricted Boltzmann Machine for Spoof Speech Detection

Neural Responses to Speech-Specific Modulations Derived from a Spectro-Temporal Filter Bank

Robust Voice Activity Detection Algorithm Based on Feature of Frequency Modulation of Harmonics and Its DSP Implementation

Contact Info

Product

Resources

About