Replication Data for: Synthetic Data is All You Need: a Voice Cloning and Text-to-Speech Pipeline for Wake Word Detection

Wake word detection classifier training dataset contains: 269992 synthetic wake word samples, generated using text to speech (TTS) and voice cloning (VC) 97846 synthetic wake word partials, i.e. 'hey' and 'snips' respectively 2171716 negative, real human samples from the TTS training or VC reference datasets all audio files are 2.4s long, single channel, 16 kHz, and stored as .flac the entire dataset is roughly 1600h