Digital filter design method and apparatus for noise suppression by spectral substraction

Abstract

A digital filter design apparatus for noise suppression by spectral subtraction includes a first spectrum estimator for determining a high frequency resolution noisy speech power spectral density estimate from a noisy speech signal block. A second spectrum estimator determines a high frequency resolution background noise power spectral density estimate from a background noise signal block. Averaging units form a piece-wise constant noisy speech power spectral density estimate and a piece-wise constant background noise power spectral density estimate. These averaging units are controlled by devices for adapting the length of individual segments to the shape of the high frequency-resolution noisy speech power spectral density estimate and for using the same segmentation in both piecewise constant estimates. A piece-wise constant digital filter transfer function is determined using spectral subtraction-based on the piece-wise constant noisy speech power spectral density estimate and the piece-wise constant background noise power spectral density estimate.

Description

technical field [0001] The invention relates to a digital filter design, especially the design of a filter in the frequency domain. Background technique [0002] In several applications, the digital filter H[k] in the frequency domain is designed "on the fly". An example is noise suppression using spectral subtraction. Another example is the design of frequency selective nonlinear processors for echo cancellation. A characteristic feature of these applications is the complexity of the filter design methodology. Since the filters are updated frequently, it places a heavy burden on the hardware / software that executes these design algorithms. [0003] Reference [1] describes a method that divides the frequency domain k into segments of equal or unequal length, and for each segment H[K] and the basic power of the speech signal contaminated by noise or echo Estimated value of spectral density Φ x [k] acts on a constant value. This reduces complexity, since the filter H[k] n...

AI Technical Summary

Problems solved by technology

However, this method also has the disadvantage that it is possible to split a peak of H[k] into two different segments

This can cause fluctuation peaks, creating annoying "musical noise"

Also reduces the sharpness of the spectrum, further degrading speech quality

Images