An Adaptive-Learning-Based Generative Adversarial Network for One-to-One Voice Conversion

Voice conversion (VC) emerged as a significant domain of research in the field speech synthesis recent years due to its emerging application voice-assistive technologies, such automated movie dubbing speech-to-singing conversion, name few. VC deals with vocal style one speaker another while keeping linguistic contents unchanged. Nowadays, generative adversarial network (GAN) models are widely used for feature mapping from source target speaker. In this article, we propose an adaptive-learning-based GAN model, called ALGAN-VC, improve one-to-one speakers. Our ALGAN-VC framework consists some approaches quality and voice similarity between We incorporate dense residual architecture into generator efficient learning also includes adaptive mechanism compute loss function proposed model. Moreover, boosted rate approach is incorporated enhance capability The model tested on Conversion Challenge 2016, 2018, 2020 datasets along our self-prepared Indian regional-language-based dataset. addition, emotional dataset considered evaluating model’s performance. objective subjective evaluations generated samples indicated that elegantly performed task by achieving high good quality.