This paper focuses on achieving high-quality, child-friendly speech generation across diverse languages and cultural backgrounds, including low-resource languages. We aim to leverage the potential of generative speech models, which have utility in practical applications such as language learning for children. To this end, we propose MultiGen, a multilingual speech generation model that utilizes an LLM architecture for speech generation tailored to low-resource languages. MultiGen aims to facilitate children's communication with AI systems in culturally appropriate contexts, using three low-resource languages: Mandarin, Malay, and Tamil with a Singaporean accent. Experimental results, including objective metrics and subjective evaluations, demonstrate that the proposed MultiGen outperforms baseline methods.
