An efficient synthesis method for sinusoidal vocoders

An FFT-based technique for synthesizing a “sum-of-sinusoids” signal is described. The technique is useful for efficient speech synthesis in sinusoidal-type vocoders. Each sinusoid to be synthesized is represented by a small number of FFT coefficients around the frequency of interest. Linear amplitude modulation of the sinusoid is approximated by convolution with a three-point sequence. Phase information is incorporated by multiplication with a phase constant. The FFT coefficients of each sinusoid, processed as above, are added together and transformed by an IFFT. From the resulting time-domain signal, an appropriate section is extracted to obtain the desired “sum-of-sinusoids” signal. When the number of sinusoids is large, the proposed technique can provide an almost 45% reduction in complexity