Building BU-TTS: A Welsh-English Voice Dataset for Modern AI

How we created a bilingual text-to-speech corpus large enough for neural networks, despite a pandemic