TTS-AGI/

mls-dacvae

	Multilingual LibriSpeech converted to DAC VAE latents




	
	Source

facebook/multilingual_librispeech

Each tar shard (~2GB) contains samples with three files per sample: {sample_key}.audio.flac # Original audio (FLAC, original sample rate) {sample_key}.dacvae.npy # DAC VAE latent [T_latent, 128] numpy float32 {sample_key}.metadata.json # All metadata + duration_seconds + chars_per_second

Model:… See the full description on the dataset page: https://huggingface.co/datasets/TTS-AGI/mls-dacvae.