This project aims to evaluate and compare recent text-to-audio generation models for synthesizing audio from natural-language descriptions, with the longer-term goal of developing personalized soundscapes that may support relaxation and better sleep. The study will investigate publicly available pretrained models and assess the quality and semantic relevance of their generated audio across controlled sets of text prompts. Evaluation will include CLAP-based text-audio similarity measures and other objective metrics, complemented by qualitative analysis. The initial experiments will provide insights into the capabilities and limitations of current text-to-audio models and their suitability for generating sleep-oriented soundscapes, providing a basis for subsequent larger-scale studies.