Fast-VGAN: Lightweight Voice Conversion with Explicit Control of F0 and Duration Parameters

TL;DR

Fast-VGAN introduces a lightweight neural network for voice conversion that allows explicit, intuitive control over pitch, duration, and speaker identity, enabling flexible and high-quality speech transformations.

Contribution

The paper presents a convolutional neural network that explicitly conditions on speech parameters for controllable voice conversion without disentanglement techniques.

Findings

Achieves high intelligibility and speaker similarity.

Provides flexible control over F0, duration, and speaker identity.

Performs well on speaker conversion and expressive speech tasks.

Abstract

Precise control over speech characteristics, such as pitch, duration, and speech rate, remains a significant challenge in the field of voice conversion. The ability to manipulate parameters like pitch and syllable rate is an important element for effective identity conversion, but can also be used independently for voice transformation, achieving goals that were historically addressed by vocoder-based methods. In this work, we explore a convolutional neural network-based approach that aims to provide means for modifying fundamental frequency (F0), phoneme sequences, intensity, and speaker identity. Rather than relying on disentanglement techniques, our model is explicitly conditioned on these factors to generate mel spectrograms, which are then converted into waveforms using a universal neural vocoder. Accordingly, during inference, F0 contours, phoneme sequences, and speaker embeddings can be freely adjusted, allowing for intuitively controlled voice transformations. We evaluate our approach on speaker conversion and expressive speech tasks using both perceptual and objective metrics. The results suggest that the proposed method offers substantial flexibility, while maintaining high intelligibility and speaker similarity.