This is a filtered version of LibriTTS-R. It has been filtered based on two sources:
LibriTTS-R paper [1], which lists samples for which speech restoration have failed
LibriTTS-P [2] list of excluded speakers for which multiple speakers have been detected.
LibriTTS-R [1] is a sound quality improved version of the LibriTTS corpus which is a multi-speaker English corpus of approximately
585 hours of read English speech at 24kHz sampling rate… See the full description on the dataset page: https://huggingface.co/datasets/parler-tts/libritts_r_filtered.
