Transcribe a sermon, khutbah or bayan in the language it was preached

Why long sermons come back shortened from a chat window

A sermon of forty-five minutes, or a Friday khutbah together with the talk some mosques give before it, holds thousands of words. Chat assistants that accept audio cap how long the file can be, free plans often cannot take audio at all, and a long transcript does not fit in one reply, so it is summarized or ends partway. For preaching in Amharic, Tigrinya, Urdu or Yoruba there is a second risk: the assistant answers in English, and you get its version of the message rather than the words spoken.

This tool cuts the recording into parts of up to ten minutes and transcribes each part completely, in the language preached, with the same instructions. Timestamps are shifted by where each part starts and the parts are joined in order. A translation, if you choose one, is made paragraph by paragraph and placed under the original, so a reader can always see what was actually said.

Who uses it

• Churches publishing the text of Sunday sermons or homilies on their website
• Mosques sharing a Friday khutbah in English for members who do not follow the Arabic or Urdu
• Diaspora congregations preaching in Amharic, Tigrinya or Yoruba, with English for younger members
• Students keeping a written record of a teacher's bayan or dharma talks
• Caption files for the video of a service, for members who are deaf or hard of hearing
• Archives of older recorded sermons that never had a written version

Respecting the words: what the AI does not do

• It adds no interpretation, theology or commentary of its own; the outline only arranges what the speaker said
• It uses the tradition's own terms the way the speaker does
• It lists a scripture reference only if the speaker says it, and never adds one from memory
• A verse quoted in the talk is translated from the words spoken, not looked up in a published edition, so check it against the edition you use
• It keeps quoted lines word for word, in the language they were spoken

Accuracy and better recordings

The transcript is made by an AI speech model. It does well on clear speech. A feed from the sound desk works better than a phone at the back of the hall; echo, music under the preaching, and the congregation responding all cause mistakes, and unclear words are marked [inaudible]. Recitation and sung passages may come out less accurately than speech, so check them against the text.

The outline is written by AI from the transcript only. Read it before you publish, and check names, places and references. The translation is a draft made by AI, not an authorized translation, and the transcript is not a certified record.