Sachin Singh

Subjective and Objective Analysis of Speech Enhancement Algorithms for Single Channel Speech Patterns of Indian and English Languages

ABSTRACT This paper presents a comparative study among the seven single channel speech enhancement techniques such as spectral subtraction, Wiener filtering, minimum mean square error under speech presence uncertainty (MMSE-SPU), p-MMSE, log-MMSE and modulation channel selection (MCS). For the investigation of the capability of these techniques, 12 different practical noises on five different language databases were used. The result was analysed based on subjective and objective measure. In subjective measure SNR, peak signal-to-noise ratio (PSNR), segmental-SNR (SSNR) and mean square error (MSE) were considered, whereas for objective measure speech the intelligibility index was taken. The different language (Hindi, Kannada, Malayalam, Bengali and English) databases were taken from the Noizeus speech corpus and IIIT-H Indic speech database, while the noise database was obtained from the Noizex-92 noise corpus. The algorithms were implemented in MATLAB. The results obtained are very encouraging and helpful in the selection of single channel speech enhancement technique for practical application-based noise reduction. Further, among all the mentioned methods, MCS shows overall better performance for the five languages and 12 different practical noises.

Single Channel Speech Enhancement for Mixed Non-stationary Noise Environments

Speech enhancement is very important step for improving quality and intelligibility of noisy speech signal. In practical environment more than one noise sources are present, hence it is necessary to design a technique/ algorithm that can remove mixed noises or more than one noises from single-channel speech signals. In this paper, a single channel speech enhancement method is proposed for reduction of mixed non-stationary noises. The proposed method is based on wavelet packet and ideal binary mask thresholding function for speech enhancement. Db10 mother wavelet packet transform is used for decomposition of speech signal in three levels. After decomposition of speech signal a binary mask threshold function is used to threshold the noisy coefficients from the noisy speech signal coefficients. The performance of the proposed wavelet with ideal mask method is compared with Wiener, Spectral Subtraction, p-MMSE, log-MMSE, Ideal channel selection, Ideal binary mask, hard and soft wavelet thresholding function in terms of PESQ, SNR improvement, Cepstral Distance, and frequency weighted segmental SNR. The proposed method has shown improved performance over conventional speech enhancement methods.

A wavelet based method for removal of highly non-stationary noises from single-channel hindi speech patterns of low input SNR

This paper presents a binary mask thresholding function in Doubachies10 wavelet transform for enhancement of highly non-stationary noise mixed single-channel Hindi speech patterns of low (negative) SNR. In the wavelet transform, a five level of decomposition is used and detailed coefficients of all five levels are given to binary mask thresholding function for removing noise and enhancing the speech patterns. The robustness of the proposed method is compared with the wildly popular methods such as log-mmse, test-psc, Wiener, IdBM, and spectral-subtraction on the basis of performance measure parameters viz SNR, PSNR, PESQ, and Cepstrum distance. The algorithms were implemented in MATLAB 7.1.

Binary mask based method for enhancement of mixed noise speech of low SNR input

Abstract This paper presents a noise reduction method based on binary mask thresholding function for enhancement in single channel speech patterns of mixed highly non-stationary noises with low (negative) input SNR. For this purpose, a mixed highly non-stationary noisy speech database is generated by using noise and clean speech database of AURORA and INDIC speech, respectively. Results are compared with widely used methods such as Daubechies13, Daubechies40, Symlet13, Coiflet5, Wiener, Spectral Subtraction, and log-MMSE for performance evaluation in terms of SNR, PESQ, and Cepstrum distance parameters. In comparison to other methods the proposed single-channel speech enhancement method shows satisfactory results and obtained significant improvement in speech quality and intelligibility.

A fuzzy mask based on wavelet packet for improving speech quality and intelligibility

In this letter, a novel fuzzy mask based on wavelet is introduced for improving speech quality and intelligibility. The proposed fuzzy mask is based on wavelets soft and hard threshold limits to retain the time frequency (T-F) units of the enhanced spectrum. In contrast to prior methods, the proposed fuzzy mask approach eliminates the need of true speech spectrum to obtain a binary mask. Speech spectrum is degraded at signal-to-noise (SNR) levels (0, 5, and 10 dB) in babble, and pink, noise environments. Objective and subjective parameters MOS and PESQ indicated significant improvements over other related and binary mask methods.

Evaluation of noise estimation techniques for single-channel speech in low SNR noise environment

This paper investigates the performance capability of noise estimation techniques for single-channel speech. These techniques are evaluated in presence of low SNR noises (i. e. f16, babble, white, and pink). The noise estimation techniques have major impact on the quality and intelligibility of denoised speech pattern. The noise estimation techniques are evaluated in frequency domain in terms of quality and intelligibility measure parameters. The Perceptual Evaluation of Speech Quality (PESQ), Weighted Spectral Slop metric (WSS), Frequency Weighted Segmental SNR (fw-SNRseg), Speech Intelligibility Index (SII), and output SNR parameters are used for performance evaluation of low SNR noises mixed speech patterns. The sampling frequency used for processing is 8000 Hz and all algorithms are implemented in MATLAB 7.14.