nineninesix/expresso-conversational-en-nano-codec-dataset

Expresso Conversational EN Nano-Codec Dataset This dataset is built upon the Expresso conversational dataset and re-encoded using NVIDIA’s NeMo Audio Codec into nano audio tokens. It is designed for fine-tuning multimodal LLMs and speech systems (TTS/ASR) that rely on codec-based audio token representations. Dataset Structure text: transcription of the utterance. speaker: speaker identifier (string). nano_layer_1 … nano_layer_4: tokenized audio… See the full description on the dataset page: https://huggingface.co/datasets/nineninesix/expresso-conversational-en-nano-codec-dataset.