Waveform generation for text-to-speech synthesis using pitch-synchronous multi-scale generative adversarial networks
This work addresses the computational bottleneck in TTS synthesis for applications requiring real-time or efficient speech generation, though it is incremental as it builds on existing GAN and WaveNet methods.
The paper tackled the slow inference problem in neural text-to-speech synthesis by exploring generative adversarial networks (GANs) for waveform generation, finding that a GAN-based glottal excitation model achieved quality and voice similarity comparable to a WaveNet vocoder.
The state-of-the-art in text-to-speech synthesis has recently improved considerably due to novel neural waveform generation methods, such as WaveNet. However, these methods suffer from their slow sequential inference process, while their parallel versions are difficult to train and even more expensive computationally. Meanwhile, generative adversarial networks (GANs) have achieved impressive results in image generation and are making their way into audio applications; parallel inference is among their lucrative properties. By adopting recent advances in GAN training techniques, this investigation studies waveform generation for TTS in two domains (speech signal and glottal excitation). Listening test results show that while direct waveform generation with GAN is still far behind WaveNet, a GAN-based glottal excitation model can achieve quality and voice similarity on par with a WaveNet vocoder.