End-to-end LPCNet: A Neural Vocoder With Fully-Differentiable LPC Estimation
This work addresses the complexity and feature constraints in neural vocoders for speech synthesis, offering a more flexible and simplified system.
The paper tackles the limitations of LPCNet, a neural vocoder that requires explicit linear prediction coefficient computation from clean speech features, by proposing an end-to-end version that learns to infer these coefficients, achieving equal or better quality without explicit analysis.
Neural vocoders have recently demonstrated high quality speech synthesis, but typically require a high computational complexity. LPCNet was proposed as a way to reduce the complexity of neural synthesis by using linear prediction (LP) to assist an autoregressive model. At inference time, LPCNet relies on the LP coefficients being explicitly computed from the input acoustic features. That makes the design of LPCNet-based systems more complicated, while adding the constraint that the input features must represent a clean speech spectrum. We propose an end-to-end version of LPCNet that lifts these limitations by learning to infer the LP coefficients from the input features in the frame rate network. Results show that the proposed end-to-end approach equals or exceeds the quality of the original LPCNet model, but without explicit LP analysis. Our open-source end-to-end model still benefits from LPCNet's low complexity, while allowing for any type of conditioning features.