An Efficient Method to Study Speech Intelligibility with Non-Harmonic Spectrum Structure

To study the necessity of speech harmonic spectrum structure for speech intelligibility, a synthesis method of speech-like signal with non-harmonic spectrum structure is proposed. The spectrum envelopes are estimated based on the short-time spectrum analysis of the original speech signal. The spectrum envelopes are re-sampled on nonharmonic frequencies, and the synthesized signal is produced by the sinusoid synthesis method. The synthesized signals are used in listening test experiments. The experimental results indicate that the synthesized signals with non-harmonic spectrum structure still have fundamental intelligibility and can be understood, while most of the characteristic information of the speaker is lost.