Trelis/libritts-mimi-tokens

libritts-mimi-tokens To learn about Trelis Enterprise Voice Services, see Trelis.com/voice-ai-services. LibriTTS-R encoded with kyutai/mimi (RVQ codec, 8 codebooks x 2,048 entries, 12.5 frames/sec). Two streams per row: codes_semantic (list[uint32], 12.5 fps, vocab 2,048) — codebook 0 only (WavLM-distilled, content-aligned). codes_all_flat (list[uint32], 100 fps, offset vocab 16,384) — all 8 codebooks interleaved per frame, with codebook k mapped to [k*2048, (k+1)*2048) so a… See the full description on the dataset page: https://huggingface.co/datasets/Trelis/libritts-mimi-tokens.