Exploring the Robustness of Text-to-Speech Synthesis Based on Diffusion Probabilistic Models to Heavily Noisy Transcriptions | Synapse