Temporal alignment algorithm for electropalatographic and acoustic signals in long utterances

The lack of temporal alignment between simultaneously recorded electropalatographic (EPG) and acoustic signal, recorded through different channels with different sampling frequencies, was detected for longer stretches of speech. The temporal length for the EPG signal was normally shorter than that for the acoustic signal. Corpus-based examination found that the length variation of the acoustic signal was as a linear correlation with the temporal length difference between the two signals. The lack of synchronicity between the articulatory and acoustic events was assumed to be attributed more to the non-synchronized clocks in audio and EPG acquisition devices than the reason that multiple marked palate signals appeared at the beginning of the signal. The temporal alignment algorithm for the two signals was proposed and the Matlab program was developed for serving this purpose. Two realizations of the algorithm were tested and the results were discussed regarding the usefulness in the speech research and articulatory modeling.