A uniform phase representation for the harmonic model in speech synthesis applications

Abstract

Feature-based vocoders, e.g., STRAIGHT, offer a way to manipulate the perceived characteristics of the speech signal in speech transformation and synthesis. For the harmonic model, which provide excellent perceived quality, features for the amplitude parameters already exist (e.g., Line Spectral Frequencies (LSF), Mel-Frequency Cepstral Coefficients (MFCC)). However, because of the wrapping of the phase parameters, phase features are more difficult to design. To randomize the phase of the harmonic model during synthesis, a voicing feature is commonly used, which distinguishes voiced and unvoiced segments. However, voice production allows smooth transitions between voiced/unvoiced states which makes voicing segmentation sometimes tricky to estimate. In this article, two-phase features are suggested to represent the phase of the harmonic model in a uniform way, without voicing decision. The synthesis quality of the resulting vocoder has been evaluated, using subjective listening tests, in the context of resynthesis, pitch scaling, and Hidden Markov Model (HMM)-based synthesis. The experiments show that the suggested signal model is comparable to STRAIGHT or even better in some scenarios. They also reveal some limitations of the harmonic framework itself in the case of high fundamental frequencies.

Harmonic Phase Estimation in Single-Channel Speech Enhancement Using Phase Decomposition and SNR Information

“…4, the clean spectral phase exhibits a low variance with a visible harmonic structure compared to the large variance in the noisy speech. In particular, as reported in [25], at voiced frames, the shape of the glottal pulse changes smoothly hence a low variance of phase is observed.…”

“…The harmonic structure in the clean spectral phase across time or frequency as well as across harmonics at voiced frames captured by the low variance of phase [25], inspires us to propose a time-frequency smoothing filtering approach and apply it at speech harmonics at least for voiced speech segments in order to obtain enhanced phase estimates at harmonics. In this work, we propose a method to smooth the harmonic phase across time and frequency to reduce the variance of the noisy phase at the signal harmonics.…”

“…For phase decomposition purpose, we consider the pitch-synchronous signal segmentation proposed in [28] and reported as a favorable choice for speech analysis/synthesis using a harmonic model [25]:…”

Analysis of Phase Spectrum of Speech Signals Using Allpass Modeling

“…One of the issues with current day speech synthesizers is generation of excitation source signal, which is mainly manifested in the phase spectrum of the speech signal. Attempts to incorporate phase of speech signals into synthesis systems include usage of complex cepstrum [66], adding instantaneous phase randomness features to HMM based synthesis [67] and compensating phase mismatches in concatenative synthesis [68]. Alternatively, the proposed AP filter, which models the phase spectral characteristics of speech signal, can be used for speech synthesis.…”

Initial investigation of speech synthesis based on complex-valued neural networks

“…relative phase shift [7], group delay [8], phase dispersion [9], phase distortion [10] and the complex cepstrum [6] for speech synthesis. For example, in [6] and [11], complex cepstra or a cepstrum-like representation calculated from the standard deviation of phase distortion have been modelled, respectively, using an additional independent stream in HMM-based statistical parametric speech synthesis (SPSS) to improve the quality of the vocoded speech.…”