Datasetpublic

giannisan/gemma-csm-shards

gemma-csm curated shards Encoded training shards for the gemma-csm project (Mimi RVQ codes + text). Derived from public corpora: LJSpeech (public domain), LibriTTS (CC BY 4.0), OpenS2S / CASIA-LM. These are compressed code representations, not raw audio. shards : LJSpeech, single voice (tag [lj]), short clips shards_long_lj : LJSpeech concatenated to ~16-20s long-format segments shards_libritts / shards_libritts_o : LibriTTS multispeaker shards_long :… See the full description on the dataset page: https://huggingface.co/datasets/giannisan/gemma-csm-shards.

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes194downloads

Dataset Card

gemma-csm curated shards

Encoded training shards for the gemma-csm project (Mimi RVQ codes + text). Derived from public corpora: LJSpeech (public domain), LibriTTS (CC BY 4.0), OpenS2S / CASIA-LM. These are compressed code representations, not raw audio.

—shards : LJSpeech, single voice (tag [lj]), short clips
—shardslonglj : LJSpeech concatenated to ~16-20s long-format segments
—shardslibritts / shardslibritts_o : LibriTTS multispeaker
—shards_long : LibriTTS concatenated ~22s long-context segments
—shards_opens2s : OpenS2S English query/response pairs (audio-in -> audio-out)
—shards_val : held-out validation