MelGlow: Efficient Waveform Generative Network Based on Location-Variable Convolution
This work addresses the problem of high parameter counts in neural vocoders for speech synthesis, offering a more efficient alternative to WaveNet-like architectures.
This paper proposes MelGlow, an efficient waveform generative network that uses location-variable convolutions to model waveform dependencies. It achieves better performance than WaveGlow on the LJSpeech dataset with smaller model sizes.
Recent neural vocoders usually use a WaveNet-like network to capture the long-term dependencies of the waveform, but a large number of parameters are required to obtain good modeling capabilities. In this paper, an efficient network, named location-variable convolution, is proposed to model the dependencies of waveforms. Different from the use of unified convolution kernels in WaveNet to capture the dependencies of arbitrary waveforms, location-variable convolutions utilizes a kernel predictor to generate multiple sets of convolution kernels based on the mel-spectrum, where each set of convolution kernels is used to perform convolution operations on the associated waveform intervals. Combining WaveGlow and location-variable convolutions, an efficient vocoder, named MelGlow, is designed. Experiments on the LJSpeech dataset show that MelGlow achieves better performance than WaveGlow at small model sizes, which verifies the effectiveness and potential optimization space of location-variable convolutions.