~/projects/byu-speeches
BYU Speeches, Word by Word
What is said at Brigham Young University's devotionals and forums, by whom, and how has it changed? Every speech in the BYU Speeches archive, 2,546 of them from 1891 to 2026, 1,975 with a transcript totalling 8.0M words, by 1,083 speakers, in one searchable database, with a programme of studies of the text and of the audio, including what makes the Marriott Center laugh. General Conference has a database of its own.
By event: Devotional 2,051 · Commencement 220 · Forum 161 · University Conference 114. For 557 speeches, mostly from the 1950s and 1960s, the archive has a recording but no transcript; they are listed with their audio and left out of the text measures. Most years before the 1950s have a handful of speeches.
Studies
The studies planned for this corpus, in the order they will ship. Two are possible only here: laughter and applause from the audio, and reception against the speaker's draw. Studies that compare the two corpora live on one page linked from both hubs.
Guess the year
A model that dates a talk from its words alone, and the talks it gets most wrong: the ones ahead of or behind their time.
How phrases spread
New coinages such as “covenant path”, who adopted them and how fast, speaker by speaker and year by year.
Who quotes whom
Near-duplicate passages across every talk: recycled stories, self-quotation and how long a passage stays in circulation.
Who gets quoted
Every quotation classified as scripture, a prior leader, a secular author or personal, and the share of each over time.
What a paragraph does
Paragraphs tagged as doctrine, policy, exhortation, story or testimony, and the mix by era, calling and gender.
Who the talk is for
Predicting the session from the content, and where women's and priesthood sessions have converged.
What rose and what faded
Topic models by decade: the doctrines that rose, those that quietly faded, and whether BYU and conference drifted together.
How words changed meaning
Embeddings trained per era to measure how words like priesthood, family, gather and covenant moved.
Style fingerprints
Clustering speakers by style, and whether apostles' style converges after ordination.
The stories they tell
Narrative passages extracted, embedded and clustered into story families, from pioneer stories to global-member stories.
The shape of a talk
Whether a standard structure of a talk emerged, and when it hardened.
Whose talk is this
Predicting the speaker, then the calling, from the text, and which speakers the model confuses.
What outperforms the speaker
Predicting which speeches made the site's all-time popular list from their words, controlling for calling, event and year, to find the content that beats its draw.
What makes a hall laugh
Laughter and applause detected in the audio, aligned to the transcript, and a model of what earns them.
How they sound
Pace, pauses and pitch range clustered across speakers, and whether voice clusters match text clusters.
What only Latter-day Saints say
Vocabulary that marks the Latter-day Saint pulpit against papal encyclicals, Spurgeon, Wesley and the presidents' inaugurals, and BYU against conference.
Every speech
Search and filter by year, event, speaker, topic, title and the scripture a speech cites most. Open any speech for its length, reading grade and citations, with links to the transcript, audio and video at speeches.byu.edu, or open a speaker for their whole record.
Sources & method
Every record comes from speeches.byu.edu: the transcript, title and BYU's event and topic tags from the site's public WordPress API, and the delivery date, the speaker's position line, the audio file and the video from each speech's page. A speech whose page carries under a hundred words of text is treated as having no transcript. Speakers are matched to the leadership database by name. The speeches are copyrighted by Brigham Young University; this site stores the full text offline for analysis and publishes only metadata, aggregate measures and a single opening sentence per speech, with a link to the official page for the rest.
Word counts exclude headings and the notes that follow a transcript. The reading grade is Flesch–Kincaid, computed on the prose. Scripture citations are references written out in the text with a chapter and verse.