This paper describes the National Research Council of Canada's (NRCC) submission to the 2025 Blizzard Challenge involving building text-to-speech systems for the Bildts language. I describe the data and processing, training, and inference procedures for the submitted system, and present my interpretation of the evaluation. Using a combination of Dutch and Bildts data, I trained a custom grapheme-to-phoneme engine and a StyleTTS2 based text-to-speech model. I also outline an approach for selecting reference audio for style vector computation at inference, and provide publicly accessible audio samples. Overall, the NRCC submitted system ranked among the top systems submitted to the challenge. It is identified as system G in the evaluation results.