AAS-VC: On the Generalization Ability of Automatic Alignment Search based Non-autoregressive Sequence-to-sequence Voice Conversion

TL;DR

This paper introduces AAS-VC, a non-autoregressive voice conversion model that improves generalization on small datasets by eliminating reliance on external duration models, demonstrating superior performance with only 5 minutes of training data.

Contribution

The paper proposes AAS-VC, a novel non-AR seq2seq voice conversion model using automatic alignment search to enhance generalization on limited data.

Findings

AAS-VC outperforms traditional models on small datasets.

Eliminating external duration dependency improves generalization.

AAS-VC achieves effective voice conversion with only 5 minutes of training data.

Abstract

Non-autoregressive (non-AR) sequence-to-seqeunce (seq2seq) models for voice conversion (VC) is attractive in its ability to effectively model the temporal structure while enjoying boosted intelligibility and fast inference thanks to non-AR modeling. However, the dependency of current non-AR seq2seq VC models on ground truth durations extracted from an external AR model greatly limits its generalization ability to smaller training datasets. In this paper, we first demonstrate the above-mentioned problem by varying the training data size. Then, we present AAS-VC, a non-AR seq2seq VC model based on automatic alignment search (AAS), which removes the dependency on external durations and serves as a proper inductive bias to provide the required generalization ability for small datasets. Experimental results show that AAS-VC can generalize better to a training dataset of only 5 minutes. We also conducted ablation studies to justify several model design choices. The audio samples and implementation are available online.