Modern text-to-speech (TTS) systems generate high-quality natural-sounding speech, but they only support a limited number of languages. Building data-hungry systems that require large amounts of accurately paired speech and text is challenging for languages with limited resources. Yiddish is a minority language that lacks many of the computational resources available in more widely-spoken languages. No modern TTS system exists for Yiddish. We introduce the Reading Electronic Yiddish Documents or REYD (Yiddish for 'speech') project. Found data is used to create a high-quality, hand-corrected TTS dataset. This dataset is used to train FastSpeech2, a state-of-the-art TTS system. A formal evaluation by expert and non-expert listeners found that the system produced speech that was both intelligible and natural-sounding. The results of this evaluation were used to further improve the dataset. The final hand-corrected dataset, code for creating a TTS system, trained models and other Yiddish text processing tools used in our work are publicly released. We hope the availability of these resources will enable new speech technology projects that better serve the needs of Yiddish-speaking communities.