MFCC Compensation for Improved Recognition of Filtered and Band-Limited Speech

Abstract

This paper addresses the problem of bandwidth expansion for the purpose of robust speech recognition. We show that an HMM-based ASR engine trained with full spectrum range data (0-8kHz) can successfully perform speech recognition tasks over band-filtered test data compensated by means of a series of simple MFCC parameter corrector functions. The problem is important when ASR is employed for audio streams of unknown frequency bandwidth common in spoken document retrieval. Evaluation is based on recognition rates. Accuracy varies depending on the width and spectral regions eliminated, but the system shows great advantages over the use of uncompensated filtered test data. The theoretical maximum recognition rates using corrector functions over filtered test data are very close to the base rate (unfiltered data) even when the greatest part of the spectrum of the original data is suppressed. These rates are even better than those obtained in the matched train/test HMMs with filtered data.

A Review on Speech Recognition Technique

“…This technique is used to mainly analyze and extract the characteristic of the excitation state. [8].…”

“…The main disadvantage of statistical models is that they must take priori modeling assumptions which are answerable to be inaccurate, handicapping the system performance. In recent years, a new approach to the challenging problem of conversational speech recognition has emerged, holding a promise to overcome some fundamental limitations of the conventional Hidden Markov Model (HMM) approach (Bridle et al, 1998 [8]; Ma and Deng, 2004 [9]). This new approach is a radical departure from the current HMM-based statistical modeling approaches.…”

Statistical class-based MFCC enhancement of filtered and band-limited speech for robust ASR

“…The algorithm presented in this paper is an evolution of our previous work on polynomial corrector functions as a means of compensation for MFCC parameters [8]. Stereo data (fulland restricted-bandwidth versions of train files) are used to obtain polynomial functions that map the restricted-bandwidth realizations to those in the full-bandwidth space; we call them Polynomial Corrector Functions (PCFs).…”

Bandwidth extension of narrowband speech in log spectra domain using neural network

“…The expansion of the spectral envelope is a more challenging task and strictly depends on the features that estimate the spectral envelope. In the BWE procedure, spectral envelope information is usually represented as a set of cepstral coefficients [2], linear predictive coding (LPC) coefficients [4], line spectral frequency coefficients [5][6][7], mel-frequency cepstral coefficients (MFCCs) [8][9][10], a set of autocorrelation coefficients [11], or melspectrum coefficients [12]. The codebook technique is the fundamental envelope prediction method.…”