Duration modelling and evaluation for Arabic statistical parametric speech synthesis

Abstract

Sound duration is responsible for rhythm and speech rate. Furthermore, in some languages phoneme length is an important phonetic and prosodic factor. For example, in Arabic, gemination and vowel quantity are two important characteristics of the language. Therefore, accurate duration modelling is crucial for Arabic TTS systems. This paper is interested in improving the modelling of phone duration for Arabic statistical parametric speech synthesis using DNN-based models. In fact, since a few years, DNN have been frequently used for parametric speech synthesis, instead of HMM. Therefore, several variants of DNN-based duration models for Arabic are investigated. The novelty consists in training a specific DNN model for each class of sounds, i.e. short vowels, long vowels, simple consonants and geminated consonants. The main idea behind this choice is the improvement that we already achieved in the quality of Arabic parametric speech synthesis by the introduction of two specific features of Arabic, i.e. gemination and vowel quantity into the standard HTS feature set. Both objective and subjective evaluations show that using a specific model for each class of sounds leads to a more accurate modelling of the phone duration in Arabic parametric speech synthesis, outperforming the state-of-the-art duration modelling systems.

Advancements in Arabic Text-to-Speech Systems: A 22-Year Literature Review

“…Table XII lists the techniques used to evaluate the studies. The subjective evaluation method was commonly applied to evaluate TTS systems, and the most commonly used method was the Mean Opinion Score (MOS) test [88], [92], [93], [94], [99], [103], [104], [105], [61], [106], [108], [110], [112], [114], [115], [117], [118], [119]. Categorial estimation (CE) tests, preference test [102], [108], [117], DMOS test [108], and DRT tests [87], [96], [116] have also been used as subjective evaluation tests in some studies.…”

“…The subjective evaluation method was commonly applied to evaluate TTS systems, and the most commonly used method was the Mean Opinion Score (MOS) test [88], [92], [93], [94], [99], [103], [104], [105], [61], [106], [108], [110], [112], [114], [115], [117], [118], [119]. Categorial estimation (CE) tests, preference test [102], [108], [117], DMOS test [108], and DRT tests [87], [96], [116] have also been used as subjective evaluation tests in some studies. Most studies measured intelligibility [91], [94], [95], [96], [99], [103], [104], [105], [61], [106], [109], [110], [112], [115], [116], [118], [120] and naturalness [94], [95], [99], [101],…”

“…Categorial estimation (CE) tests, preference test [102], [108], [117], DMOS test [108], and DRT tests [87], [96], [116] have also been used as subjective evaluation tests in some studies. Most studies measured intelligibility [91], [94], [95], [96], [99], [103], [104], [105], [61], [106], [109], [110], [112], [115], [116], [118], [120] and naturalness [94], [95], [99], [101], [103], [104], [105], [108], [109], [112], [115], [116], [118] whereas others measured pronunciation [95], [109], sound quality [95], [109], [111], prosody [91], nasality [87], graveness [87], compactness [87], clearness ...…”

“…Some studies were evaluated on words, triphones, and diphones [93], some were evaluated on both words and sentence synthesis [109], [112] while others were evaluated on one of them. Objective tests that were commonly used to evaluate include Visual Perceptual Test [96], PESQ test [61], [110], [116], [119], Mel cepstral distortion (MCD) [98], [99], [102], [106] and Root Mean Squared Error (RMSE) of energy, pitch, and duration models [99], [106], [108], [117]. Some studies have also compared their TTS systems with other existing TTS systems for a more objective evaluation [87], [91], [96], [103], [106], [111], [114], [117].…”

“…Objective tests that were commonly used to evaluate include Visual Perceptual Test [96], PESQ test [61], [110], [116], [119], Mel cepstral distortion (MCD) [98], [99], [102], [106] and Root Mean Squared Error (RMSE) of energy, pitch, and duration models [99], [106], [108], [117]. Some studies have also compared their TTS systems with other existing TTS systems for a more objective evaluation [87], [91], [96], [103], [106], [111], [114], [117]. Some studies have also evaluated their TTS systems based on a success rate formula [100], [109], [120].…”