Pitch synchronous residual excited speech reconstruction on the MFCC

Practical applications of speech recognition and dialogue systems bring sometimes a requirement to synthesize or reconstruct the speech from the saved or transmitted mel-frequency cepstral coefficients (MFCCs). Presented paper describes several approaches to the speech reconstruction based on the MFCC parameterization. Approaches differ mainly in their various possible excitations. Let us mention that for designing a MFCC reconstruction module we applied some principles usually used in speech recognition and synthesis process. We suppose the speech reconstructor together with speech synthesizer and recognizer to be a part of a speech dialogue system developed in our department.