scotus-unaligned

Larry SCOTUS Oral Argument Unaligned Materialization

Built from the URL manifest at thelarryproject/scotus-manifest.

One row per oral argument session of the U.S. Supreme Court, October Terms 2010–2025, materializing the official audio recording and transcript published at supremecourt.gov. Contains ~1,215 hours of audio un-pruned.

Some amount of work has been performed to format the PDF text into a transcript consisting of a series of speaker turns.

Fields

• docket: docket number; rearguments and multi-session arguments carry suffixes (rearg, q1, q2)
• term: October Term the argument belongs to
• date_argued: ISO date of the argument session
• page_url: url for transcript and mp3 for a term and docket, in the format https://www.supremecourt.gov/oral_arguments/audio/{term}/{docket}
• audio: Arrow audio typed column consisting of (bytes, path), where path is the mp3_url from the manifest
• pdf_url: full url to the transcript pdf file as reported on the page_url
• pdf_bytes: the full downloaded pdf in raw bytes
• transcript: the full decoded text, reformatted into: JUSTICE GINSBURG: Dialog. MR. McDAVIS: The -- the dialog -- it is dialog. JUSTICE THOMAS: Some bs (Laughter.)
• duration_s: the duration of the audio in seconds

Construction

See provenance.json for the producing commit, timestamp, and ancestry.

Known corrections

From the manifest

• 13-7120rearg: the court's page links the original argument's transcript and date this has been manually corrected to the reargument transcript and 2015-04-20, found via the year-in-review page at https://www.supremecourt.gov/oral_arguments/argument_transcript/2014

From this dataset

• Some mistakes were found in the original transcripts and have been corrected for, such as:
• minor misplacement of speaker labels

Known gaps

• Left to downstream consumers (segmenters/aligners):
• Normalization of speaker labels/ids
• Removal of inline stage directions ("(Laughter)" within a line of dialog, rather than as its own line break)
• Dropped/unmatched parentheticals are distributed as-is

The audio and transcripts are Public Domain, but this redistribution is CC-BY-4.0 under the larry project.

Preprocessing pipeline code is at https://github.com/Mekadrom/larry, and this dataset is fully reproducible with a single command in that repo:

uv run preprocessing --config_file configs/preprocessing/voice/unaligned/thelarryproject_scotus-unaligned.yaml --output_dir ~/.cache/larry/datasets/voice/unaligned/thelarryproject_scotus-unaligned --clean