Modification of Velvet Noise for Speech Waveform Generation by Using Vocoder-Based Speech Synthesizer

Abstract

This paper introduces a new noise generation algorithm for vocoder-based speech waveform generation. White noise is generally used for generating an aperiodic component. Since short-term white noise includes a zero-frequency component (ZFC) and inaudible components below 20 Hz, they are reduced in advance when synthesizing. We propose a new noise generation algorithm based on that for velvet noise to overcome the problem. The objective evaluation demonstrated that the proposed algorithm can reduce the unwanted components.

A practical method of generating whisper voice: Development of phantom silhouette method and its improvement

“…Condition 4 was a pair of two pseudo-whisper generated by P.S.2, with white noise and modified velvet noise as the drive sound source. Since the adoption of modified velvet noise is currently being considered for WORLD [6], we also investigated its effect on the phantom silhouette method.…”

Colours of velvet noise

“…Spectrally, VN is known to sound similar to white noise [1, 2], although a theoretical proof that its power spectral density (PSD) is constant has not been reported previously. So called crushed variants of VN have also been proposed, in which an unequal probability for the sign of non‐zero samples yields coloured spectra, especially with suppressed low frequencies [9, 11]. Also, it was experimentally observed that for some particular VN sequences, changing the sign probability does not affect the spectrum at all [11].…”

“…Convolution of any other signal with VN can be implemented efficiently, when the sparsity of VN is accounted for [1,3]. Several signal processing applications based on VN have evolved, such as artificial reverberation [1,[3][4][5], audio decorrelation [6,7], speech synthesis [8,9], and acoustic measurements [10].…”