Datasets:

pennydata
/

librivox-transcripts

license: cc0-1.0
language:
  - en
task_categories:
  - automatic-speech-recognition
  - text-generation
pretty_name: LibriVox public-domain chapter transcripts (9 chapters)
size_categories:
  - n<1K

LibriVox public-domain chapter transcripts

Machine-made transcripts with segment timestamps for 9 short LibriVox chapters (about 80 minutes of audio, about 14,100 words).

Files

librivox_transcripts.jsonl: one row per chapter, with segments as a list of {start, end, text} (seconds).
librivox_transcripts.csv: same rows; segments_json holds the segments as a JSON string.

Columns: title, author, chapter, source_url (direct archive.org audio file), archive_item (archive.org page), duration_seconds, language, word_count, transcript, segments.

Source

Audio: LibriVox volunteer recordings hosted on archive.org. LibriVox recordings are dedicated to the public domain, and the books read are public domain. Items:

https://archive.org/details/adventures_pinocchio_librivox (chapters 1-4)
https://archive.org/details/art_of_war_librivox (parts 1-2 and 3-4)
https://archive.org/details/alice_in_wonderland_librivox (chapter 1)
https://archive.org/details/secret_garden_librivox (chapter 1)
https://archive.org/details/peter_pan_0707_librivox (chapter 1)

License

Audio: public domain (LibriVox). Transcripts and timestamps in this dataset: CC0 1.0.

Method

Each file was fetched from archive.org and transcribed with the Apify actor audio-transcriber (language en, segments on). The text is raw machine output: it was not proofread, so expect errors in names, punctuation and rare words, and compare it with the printed book before using it as ground truth.

I built the actor and earn from its runs. Transcripts were generated with AI (speech recognition) and the dataset was assembled with AI assistance (Claude).