Assessing telephone connection speech quality involves classifying observation sequences using hidden Markov models, outputting model with maximum total emission probability

摘要

The method involves comparing the original (reference) speech with a corresp. part of the transmitted speech. The digitized speech signal of the reference statement and the digitized speech signal of the transmitted statement are subjected to short-duration spectral analysis; the successive short-duration spectra are smoothed and subjected to energy normalization; the characteristic sequences of the reference and transmitted statements are brought into coincidence by synchronization and subtracted; and the resulting observation sequences are classified using several hidden Markov models, the model with the maximum total emission probability is determined and output as the assessment result.