Mel-frequency cepstral coefficient-based bandwidth extension of narrowband speech

Abstract

Abstract We present a novel MFCC-based scheme for the BandwidthExtension (BWE) of narrowband speech. BWE is based onthe assumption that narrowband speech (0.3–3.4 kHz) cor-relates closely with the highband signal (3.4–7 kHz), en-abling estimation of the highband frequency content given thenarrow band. While BWE schemes have traditionally usedLP-based parametrizations, our recent work has shown thatMFCC parametrization results in higher correlation betweenboth bands reaching twice that using LSFs. By employinghigh-resolution IDCT of highband MFCCs obtained from nar-rowband MFCCs by statistical estimation, we achieve high-quality highband power spectra from which the time-domainspeech signal can be reconstructed. Implementing this schemefor BWE translates the higher correlation advantage of MFCCsinto BWE performance superior to that obtained using LSFs,as shown by improvements in log-spectral distortion as well asItakura-based measures (the latter improving by up to 13%). Index Terms : Bandwidth extension, high-resolution IDCT,highband certainty, mutual information, source-filter model