Improved Reproduction of Stops in Noise Reduction Systems with Adaptive Windows and Nonstationarity Detection

Abstract

A new block-based noise reduction system is proposed which focuses on the preservation of transient sounds like stops or speech onsets. The power level of consonants has been shown to be important for speech intelligibility. In single-channel noise reduction systems, however, these sounds are frequently severely attenuated. The main reasons for this are an insufficient temporal resolution of transient sounds and a delayed tracking of important control parameters. The key idea of the proposed system is the detection of non-stationary input data. Depending on that decision, a pair of spectral analysis-synthesis windows is selected which either provides high temporal or high spectral resolution. Furthermore, the decision-directed approach for the estimation of the a priori SNR is modified so that speech onsets are tracked more quickly without sacrificing performance in stationary signal regions. The proposed solution shows significant improvements in the preservation of stops with an overall system delay (input-output, excluding group delay of noise reduction filter) of only 10 milliseconds.

Robust Speech-Distortion Weighted Interframe Wiener Filters for Single-Channel Noise Reduction

“…Evaluating the ztransforms on the unit circle, the higher frequency resolution in X l (z) implies that h L [n] has a more narrow lowpass frequency response than h K [n] and that L > K. It has been shown that WOLA filter banks with longer prototype filters, such as h L [n], can still be synthesized with the same low delay as h K [n], see for instance [12], [13], [14]. The HRFB gives an improved frequency resolution representation of x[n] under the assumption that the signal is stationary over L samples.…”

Improving Performance of a Noise Reduction Algorithm by Switching the Analysis Filter Bank

“…In one implementation [15], the weighted summation of the first derivative of signal energy in four frequency regions of up to 5.5 kHz is used as a measure to mark the transient segment of the speech signal. The transient is also detected by a likelihood ratio measurement of statistical similarity of two block of data to decide whether the signal has stayed stationary or its statistical characteristics have changed, hence signifying a transient event [10].…”

“…Tests based on informal subjective listening suggest that higher gain values (~ 4dB) can be applied at high SNR values, however this gain needs to be reduced as the SNR deteriorates. A similar method is suggested in [17] to improve the intelligibility of speech signal in the presence of background noise and deployed in [10] for a noise reduction algorithm with a switching analysis/synthesis filter bank structure. .…”

“…Another approach is taken in [9] to mix the coefficients of two different analysis filter banks depending on transient detection of the signal, however, details of filter banks and complete reconstruction of the method is not addressed. The approach proposed in this study is somewhat similar to [10] in which a pair of adaptive analysis/synthesis filter banks with high/low temporal resolutions are switched (swapped) depending on the transient detection in the audio speech signal. The main difference in the approach presented in this study is that the system works in the time domain and the selected filter banks do not need to satisfy the full reconstruction restrictions during transient switching period.…”

Noise Estimation and Suppression Using Nonlinear Function withA PrioriSpeech Absence Probability in Speech Enhancement

“…Several recent studies have proposed noise estimation schemes for unknown noise signals [1][2][3][4][5][6][7][8][9][10][11][12][13][14]. The minimum statistics (MS) noise estimation scheme [7] is one that works well in nonstationary noisy environments.…”