yskim3271/vibravox_16k

Vibravox HF (16kHz)

This dataset is a pre-processed version of the Vibravox dataset (Cnam-LMSSC/vibravox), created from the speech_clean split.

For each sample, this version provides headset microphone audio as audio.acoustic_microphone and throat microphone audio as audio.throat_microphone. Both have been resampled to 16kHz.

Dataset Structure

Data Splits

• train
• dev
• test

Data Fields

• audio.acoustic_microphone: The path to the 16kHz WAV file (headset mic).
• audio.throat_microphone: The path to the 16kHz WAV file (throat mic).
• speaker_id: A unique identifier for the speaker.
• sentence_id: The index of the sample for a given speaker (per split).
• text: The normalized transcript of the audio.

Source Data & Citation

This dataset is derived from Vibravox. If you use this data in your work, please be sure to cite the original authors:

@inproceedings{bel-etal-2023-vibravox,
    title = "Vibravox: A Large-Scale Open-Source Dataset for Trans-Cultural Speech Transfer and Synthesis",
    author = "Bel, No{'e}  and
    Ly, Rachid  and
    Patrat, Jean-Marc  and
    Malkin, R{'e}mi  and
    Jir{'a}k, Jan  and
    Gauthier, Pierre-{'E}lie  and
    Baecque, Guillaume  and
    Avanzi, Mathieu  and
    Pellier, Denis  and
    P{'e}rennou, Georges",
    editor = "Bouamor, Houda  and
    Pino, Juan  and
    Bali, Kalika",
    booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2023",
    address = "Singapore",
    publisher = "Association for Computational Linguistics",
    url = "[https://aclanthology.org/2023.emnlp-main.274](https://aclanthology.org/2023.emnlp-main.274)",
    doi = "10.18653/v1/2023.emnlp-main.274",
    pages = "4514--4530",
}