Performance Improvement in Speech Enabled IVR Systems Using Artificial Band Width Extension: Recent Development

The existing time-consuming menu-driven IVRs have been replaced by new blends of Speech Powered Interactive Voice Response (SEIVR) Systems. When transmitted via telecom networks that use narrowband codecs, however, the quality and intelligibility of the speech signal suffers significantly. Providing wideband quality signal without much modification of the existing network infrastructure can only be possible with a novel technique of Artificial Band Width Extension (ABWE). ABWE is implemented in this paper using a QMF filter bank that band splits the input speech into LF and HF components before compressing and encoding the HF components using a novel data hiding technique. A QMF synthesis filter bank is used in the reconstruction process to create an artificial wideband speech signal. For implementation of the proposed model, a client-server approach with socket programming on a single machine has been used assuming no noise and no transmission errors. A comparative analysis has also been done to find out the root cause for degradation in performances of SEIVR systems. The proposed SEIVR model with ABWE has shown a major improvement in speech recognition accuracy and overall efficiency in simulations. Based on the simulation findings, it can be concluded that the recognition accuracy of an artificially extended NB speech signal is superior to that of an NB speech signal and comparable to that of a WB speech signal.