An Algorithm for Formant Analysis, Tracking and Modification

Formant tracking has been an intensively studied topic in speech research. In the work reported in this paper formants and their tracks are used for analyzing and modifying the spectral content of speech. The algorithm is based on linear prediction (LP) analysis, on finding the roots of the all-pole filter, and on a constraint-based mapping between poles, formants and formant tracks. Formant frequency and bandwidth information can be obtained from the poles assigned to formants. The main advantage of this method is its accuracy: the results are considerable even with minimal input and it can be improved significantly by providing additional information. With the ability to modify the extracted formant tracks and to re-synthesize speech, it allows a wide range of applications: a visual tool for educational purposes was implemented and the described method was successfully used to conduct an extensive formant analysis for Hungarian vowels.