Superposition of Functional Contours Based Prosodic Feature Extraction for Speech Processing

Speech signal analysis for the extraction of speech elements is viable in natural language applications. Rhythm, intonation, stress, and tone are the elements of prosody. These features are essential in emotional speech, speech to speech, speech recognition, and other applications. The current study attempts to extract the pitch and duration from historical Sindhi sound clips using the functional contours model’s superposition. The sampled sound clips contained the speech of 273 undergraduates living in 5 districts of the Sindhi province. Several Python libraries are available for the application of this model. We used these libraries for the extraction of prosodic data from a variety of sound units. The spoken sentences were categorically segmented into words, syllables, and phonemes. A speech analyzer investigated the acoustics of sounds with the power spectral density method. Meanwhile, a speech database was divided into parts contains words of different sizes (ranging from 1-letter to 5-letter words). The results illustrated the production of both minimum and maximum μ sound durations and pitches from the inhabitants of Khairpur and Ghotki districts, respectively. Both districts lie in the upper part of the Sindh province. In addition, the second parameter approach, observed versus obtained, was used to compare outcomes. We observed 5250 and 4850 durations and pitches, respectively.