Vowel phoneme identification in speech

Parameters correlated to the frequency distribution of voice signal energy were used to investigate the efficiency of a vowel phoneme identification method using a limited band of preestablished frequencies. The classification algorithm works in sequenced decisions based on the comparison of the parameter values with experimentally determined thresholds. A PDP 11/34 was used with voice signals filtered at 4.4 kHz sampled at 10 kHz and quantized at 12 bit. Sample segment duration was 25.6 msec. Results show a recognition ability of 95% of the vowel sounds and 90% recognition of the signal segments in the signal stability region, while at the limit of the stability zone the segment recognition drops to 86%.