Semi-automatic Segmentation and Marking of Pitch Contours for Prosodic Analysis

摘要

Prosody is used in both phonetics and speech synthesis systems in the literature. Pitch information is one of the extensively used prosodic information. This paper aims at semiautomatizing the process of pitch marking for prosodic analysis. Prosody is a suprasegmental information; therefore, it may be wiser to analyze the change in pitch over a segment of voiced speech instead of directly using the pitch calculated from a small window. In a particular voiced segment of speech, pitch may vary from low-to- high, high-to-low, or it may not vary at all. This work describes a method for automatically segmenting speech into certain regions having a continuous pitch contour and marking the nature of pitch change within those regions. Zero frequency filtering is used to segment the speech into voiced and unvoiced segments. This segment is further divided into small segments depending on a discontinuity present in the pitch contour. A height value of the pitch contour in the final segment is measured and accordingly marking is done. Now, automatic segmentation and markings are manually corrected by deleting, inserting, or shifting the segmentation boundaries and substituting the wrong markings. Automatic process is evaluated in terms of these four parameters.