A unique real-time speech decoder that operates from new perspectives

Four new techniques for real-time speech decoding have been combined to decode speech without training the device and to give speaker-independent segmentation and phoneme-like decoding. The first technique is a new time-domain process which detects the mouth resonance with high accuracy. The second process compensates for mouth-size and pitch fluctuations. The third technique provides vowel-like segmentation and consonant-like classification. Because of the combined power of these algorithms, a fourth technique can be used to form a table look-up of real-time analog and digital orthography of speech under non-ideal conditions. This system appears to be immune to co-articulation difficulties. Actual performance results are given.