Multi-Scale Spectrogram Modelling for Neural Text-to-Speech

Ammar Abbas, Bajibabu Bollepalli, Alexis Moinet, Arnaud Joly, Penny Karanasou, Peter Makarov, Simon Slangen, Sri Karlapati, Thomas Drugman

Abstract: We propose a novel Multi-Scale Spectrogram (MSS) modelling approach to synthesise speech with an improved coarse and fine-grained prosody. We present a generic multi-scale spectrogram prediction mechanism where the system first predicts coarser scale mel-spectrograms that capture the suprasegmental information in speech, and later uses these coarser scale melspectrograms to predict finer scale mel-spectrograms capturing fine-grained prosody. We present details for two specific versions of MSS called Word-level MSS and Sentence-level MSS where the scales in our system are motivated by the linguistic units. TheWord-level MSS models word, phoneme, and framelevel spectrograms while Sentence-level MSS models sentencelevel spectrogram in addition. Subjective evaluations show that Word-level MSS performs statistically significantly better compared to the baseline on two voices.