M. Sigmund (Germany) and P. Matejka (Czech Republic)
signal processing, automatic labeling, speech recognition.
In this paper, we present a new software environment for automatic labeling of spoken language. The labeling algorithm is based on the comparison of a set of speech parameters obtained either by the direct or by the cumulative method in the stable portion of each phoneme in the utterance. Finally, the utterance is automatically segmented into phonemes. The label segmentation is supported by visual and audible feedback allowing an optimal adjustment of processing parameters and comparative evaluation of results. Testing the program on Czech speech signal, the best total segment detection rate of 90 % was achieved. The developed program can be used both as an illustrative approach for teaching speech processing and a powerful tool for parameter setting in real-world applications using signal processors.
Important Links:
Go Back
