Speech Enhancement is the process of estimating the characteristics of the speech and enhancing that speech.
Noise Reduction is the process of estimating the characteristic of noise and reducing that noise.
The aim of Speech Enhancement or Noise Reduction is to improve the Intelligibility and/or Quality of Speech.
Improving Speech Intelligibility means that the words, phrases and sentences can be more clearly heard and hence understood.
Improving Speech Quality means that the words, phrases and sentences can be more clearly heard and hence understood.
Spectral SubtractionSpectral Subtraction is the process of subtracting a spectral estimate of noise from the speech plus noise spectrum to extract speech from noise. If the spectrum of noise does not change over time, like with a tone, then overall spectral subtraction can be applied. If the noise changes over time then the estimate of the noise spectrum has to vary to reflect the change. Estimates of noise can be taken using the noise present between the speech. How accurately noise is represented, determines how effective is spectral subtraction to extract intelligible speech. The artefacts of |
|---|
Scalogram EnhancementScalogram Enhancement is the process of dividing speech plus noise into a combination of simple time frequency functions then adaptively altering by frequency or amplitude. Two common types of scalogram enhancement techniques are wavelet enhancement and spectrogram enhancement that have pulse shapes (wavelets) and sine/cosine as the basic functions. Scalogram Enhancement takes advantage of short time processing to adjust to the changes in frequency and amplitudes with time. This has the benefit of inherently representing short time changes in noise. |
Statistical Filtering EnhancementStatistcal Filtering Enhancementinvolves extracting the speech signal from the speech plus noise signal using the miminisation of some statistical aspect of the signal. Wiener Filtering involves estimation of clean speech or noise suppression filter parameters by using minimisation of the mean square error between actual and estimated. Maximum Likelihood (ML), Minimum mean squared error (MMSE) and Maximum a-posteriori (MAP) statistical model methods can be used to derive the response of a noise suppression filter. .... |
Feature Extraction and EnhancementFeature Extraction and Enhancement involves extracting features of the speech signal from the speech and enhancing just those features. ... .... .... |
Predictive Technology EnhancementPredictive Technology Enhancement involves using neural nets and/or hidden markov models to enhance Speech through training. ... .... .... |
Component Analysis EnhancementComponent Analysis Enhancement involves separating a multisource speech plus noise signal into its component source signals which are then adaptively filtered..... .... .... |