Neural HMMs are all you need (for high-quality attention-free TTS)
This addresses the issue of training inefficiency and incoherent synthesis in neural TTS for speech synthesis applications, though it is an incremental improvement combining old and new paradigms.
The paper tackles the problem of attention failures in neural text-to-speech (TTS) by replacing non-monotonic attention with an autoregressive hidden Markov model (HMM) defined by a neural network, resulting in a system that learns to speak with fewer iterations and less data while achieving comparable naturalness to Tacotron 2.
Neural sequence-to-sequence TTS has achieved significantly better output quality than statistical speech synthesis using HMMs. However, neural TTS is generally not probabilistic and uses non-monotonic attention. Attention failures increase training time and can make synthesis babble incoherently. This paper describes how the old and new paradigms can be combined to obtain the advantages of both worlds, by replacing attention in neural TTS with an autoregressive left-right no-skip hidden Markov model defined by a neural network. Based on this proposal, we modify Tacotron 2 to obtain an HMM-based neural TTS model with monotonic alignment, trained to maximise the full sequence likelihood without approximation. We also describe how to combine ideas from classical and contemporary TTS for best results. The resulting example system is smaller and simpler than Tacotron 2, and learns to speak with fewer iterations and less data, whilst achieving comparable naturalness prior to the post-net. Our approach also allows easy control over speaking rate.