WavePulse Radio Raw Transcripts
WavePulse Radio Raw Transcripts is a huge dataset containing segmented transcripts of live radio streams from 396 American stations, covering debates, news, and political shows.
Over 250 million text segments from 750,000+ hours of radio, Parquet files by state and month
Apache 2.0
Description
This dataset includes over 250 million text segments from automatic transcriptions by WhisperX of 750,000+ hours of live radio recordings across the United States. The files are organized into Public Prosecutor's Office by State and by month.
What is this dataset for?
- Train and evaluate NLP models on vast textual data from speech
- Analyze political debates and radio programs with fine segmentation
- Improve diarization, speech recognition, and oral language comprehension
Can it be enriched or improved?
Yes, the dataset can be supplemented by manual annotations, summaries, or extended to other languages and periods. Additional corrections or filtering may improve the quality for some applications.
🔎 In summary
🧠 Recommended for
- Oral NLP researchers
- Speech recognition model developers
- Political analysis
🔧 Compatible tools
- Hugging Face Datasets
- PySpark
- Pandas
- NLP frameworks (SpacY, transformers)
💡 Tip
Use segment filtering methods to extract the most relevant passages according to the targeted application.
Frequently Asked Questions
What is the main format for the dataset files?
The dataset is organized into Parquet files, segmented by state and month, containing the segmented transcripts.
Does this dataset contain speaker diarization?
Yes, automatic diarization was done with WhisperX to separate and identify speakers.
Can this dataset be used for political and social analysis?
Yes, the dataset covers numerous debates, news shows, and political discussions across the United States in 2024.
