Assessing the Intelligibility and Quality of HMM-based Speech Synthesis with a Variable Degree of Articulation

This paper focuses on the assessment of both the intelligibility and the quality of speech when using a variable degree of articulation (hypo/hyperarticulation) in the framework of HMM-based speech synthesis. Intelligibility is evaluated when the synthesizer is working in adverse conditions. The adaptation of a neutral speech synthesizer to generate hypo and hyperarticulated speech is first performed. Simulated noisy and reverberant conditions are then applied to the speech produced by the latter synthesizers. The intelligibility of the resulting speech is assessed by a Semantically Unpredictable Sentences (SUS) test. Results of this test quantify how the possibility of varying the degree of articulation improves the intelligibility of synthetic speech in various adverse conditions. In a second test, natural and synthetic speech quality is evaluated through an Absolute Category Rating (ACR) test. This test allows the assessment of hypo/hyperarticulated speech through various dimensions: comprehension, pleasantness, non-monotony, naturalness, fluidity and pronunciation.