Many-to-Many Voice Conversion with Out-of-Dataset Speaker Support
This enables voice conversion for new speakers without retraining, addressing a bottleneck in practical applications, though it is incremental as it builds on Cycle-GAN methods.
The paper tackles the problem of many-to-many voice conversion for speakers not in the training set, achieving good style conversion for out-of-dataset speakers without requiring re-training, with subjective tests showing comparable quality to state-of-the-art for in-dataset speakers.
We present a Cycle-GAN based many-to-many voice conversion method that can convert between speakers that are not in the training set. This property is enabled through speaker embeddings generated by a neural network that is jointly trained with the Cycle-GAN. In contrast to prior work in this domain, our method enables conversion between an out-of-dataset speaker and a target speaker in either direction and does not require re-training. Out-of-dataset speaker conversion quality is evaluated using an independently trained speaker identification model, and shows good style conversion characteristics for previously unheard speakers. Subjective tests on human listeners show style conversion quality for in-dataset speakers is comparable to the state-of-the-art baseline model.