←

Dashboard for Twitch Transcripts

TLDR

  • I transcribed a week of HasanAbi's Twitch streams, from the archived video audio, with a speech recognition model.
  • I counted how much of each stream had speech and which words came up most.
  • I found that three Michigan politicians were named most often in the final streams before the midterms.
  • The panels below show hours, topics and words by stream. Methods are at the end.

Show

Stream length Speech

Topic

Methods. The streams are the eight HasanAbi Twitch broadcasts from October 26 to November 2, 2026, one a day. Dates are stream start dates in Pacific time. The audio of each archived stream was transcribed in full with WhisperX, using the Whisper large-v3 model. Speech hours are the summed length of the segments where the model wrote text. Words are counted as the model wrote them, split on spaces. The transcripts include every voice on stream: the host, guests, and audio from videos played on stream. Topic and word rates count word tokens, which are runs of letters with an optional apostrophe, per 10,000 tokens in the stream. A possessive counts as its word, so "Israel's" counts as "Israel". Each topic is a fixed word list, shown under its chart. The word grid opens on word pairs. A pair is two tokens spoken one after the other in the same segment. Neither token is a common English word, and each has at least three letters. Pairs with the streamer's own name were removed. The single-words view uses the 12 most frequent words over all eight streams after common English words were removed. Spoken filler that the common-word list misses was also removed, for example "thank" and "wait". So were first names whose surname is already counted, and the streamer's own names. Speech recognition makes mistakes, mostly in names and in crosstalk, so counts are approximate. This page shows only counts and rates; no transcript text is published.

References

HasanAbi. (2026). Twitch past broadcasts, October 26 to November 2, 2026 [Video recordings]. Twitch. https://www.twitch.tv/hasanabi

Bain, M., Huh, J., Han, T., & Zisserman, A. (2023). WhisperX: Time-accurate speech transcription of long-form audio. Proceedings of Interspeech 2023, 4489 to 4493. https://github.com/m-bain/whisperX

Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., & Sutskever, I. (2023). Robust speech recognition via large-scale weak supervision. Proceedings of the 40th International Conference on Machine Learning. https://github.com/openai/whisper

Bostock, M. (2023). D3.js (Version 7) [Computer software]. https://d3js.org/