A pitch-synchronous analysis/synthesis system to independently modify formant frequencies and bandwidths for voiced speech

Abstract Based on the covariance method, we have developed an analysis/synthesis system which is capable of independent manipulation of the formant frequencies and bandwidths for voiced speech. The analysis is performed pitch synchronously and is based on the local minimum of the normalized squared error. Once the formant frequencies and their bandwidths have been estimated, modifications are performed by altering the predictor coefficients, so that the modified formants and/or bandwidths are the solution to the new polynomial equation. The input to the synthesis filter is the residual signal and a trapezoidal time window is utilized to eliminate the buzz of the output speech. This system has applications in voice modification and speech perception as a tool for investigating voice quality and personality.