BLURB-synth: Synthetic audio data based on BLURB corpora

Dataset Summary

Synthetic audio data based on BLURB corpora. More details coming soon...

Supported Tasks and Leaderboards

• Biomedical Language Understanding and Reasoning Benchmark (BLURB)
• Text-to-Speech
• Automatic-Speech-Recognition

Languages

English

Data Structure

Data Instances

Coming soon...

Data Fields

Coming soon...

Data Subsets

Data is divided in two different subsets:

• audio: This includes metadata such as clip length, column and chunk (each sample generated one or more chunks of less than 30 seconds), and audio data as bytes stored in Parquet files (folder clips).
• transcripts: Transcription data with pre-calculated WER and M-WER (see paper). Transcriptions are provided for the entire text and not by individual chunks of audio.

Data Splits

Two splits are provided:

• single with single-speaker audio and transcriptions.
• multi with multi-speaker audio and transcriptions.

Dataset Creation

Curation Rationale

Source textual data comes form the test split of the following BLURB corpora:

• BC2GM
• BC5CDR-chem
• BC5CDR-disease
• JNLPBA
• NCBI-disease
• BIOSSES
• HoC
• BioASQ
• PubMedQA

Single-speaker data was generated for the full contents of the test split, while multi-speaker data is provided only for 100 randomly-selected rows from each corpora.

The audio data in this dataset has been generated with three different text-to-speech systems as described in our paper.

Automatic transcriptions (columns *_pred) were generated using Whisper-Medium.en as explained in our paper. Use the value of WER to identify the quality of each transcription relative to the original text (WER == 0 indicates and excellent match).

Source Data

BLURB textual data was downloaded from michiyasunaga/LinkBERT. We thank the developers of michiyasunaga/LinkBERT for making the pre-processed BLURB data available to the community.

Personal and Sensitive Information

The dataset does not contain any sensitive information. BLURB source textual data has been downloaded from public repositories.

Additional Information

Dataset Curators

• Ricardo Rodriguez (LIA, Avignon University)
• Stéphane Huet (LIA, Avignon University)
• Benoit Favre (LIS, Aix-Marseille Université)
• Mickael Rouvier (LIA, Avignon Université)

This work was supported by funds from the ANR (French National Agency for Research), for the project ANR MALADES.

Licensing Information

This work is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). See LICENSE.txt for the full text of the deed. To view an online copy of this license, visit https://creativecommons.org/licenses/by/4.0/.

Citation Information

Publication in process.

DOI: ...