NEWVectors or files. Pick a path.Start →
    Search & Discovery
    8 min read
    Updated 2026-09-30

    How Do I Find Where Someone Said Something in Our Meeting Recordings?

    You remember a decision, a number or a promise from a call, and there are hundreds of recordings. This explains how to search them by what was said, why exact-word search misses so much of it, how to find the moment in a long meeting, and how to search what was shown on screen as well as what was spoken.

    Meeting Recordings
    Transcription
    Video Search
    Audio Search
    Knowledge Management

    How do I find where someone said something in our meeting recordings?



    Transcribe the recordings with timestamps, index the transcripts in short passages, and search them two ways at once: by exact words for names and numbers, and by meaning for everything else. Each result should point to the meeting and the second the passage starts, so you can jump straight to it. If slides or documents were shared on screen, read the text from those frames too, because the answer is often on a slide nobody read aloud.

    Why can't I just search the transcripts?



    You can, and it is the right first step, but plain transcript search has three gaps.

  1. It only finds the words that were said. Someone asking "when did we agree on the price" will not
  2. find a passage that says "let's go with forty-five a seat".
  3. It works one meeting at a time in most meeting tools, so a question that spans a quarter of calls
  4. means opening each one.
  5. Names, product terms and numbers are where transcription errs most. A customer's name or an
  6. internal code word can be transcribed several different ways, and exact search misses every variant but the one you typed.

    What are the options?



    ApproachFindsMissesEffort
    Search inside the meeting tool's transcriptExact words in one meetingParaphrases; anything across many meetingsNone if transcription is on
    Export transcripts and search them togetherExact words across every meetingParaphrases and misspelled namesLow
    Semantic search over timestamped transcript passagesPassages by meaning, across meetings, with the momentExact codes and numbers, unless combined with keyword searchModerate
    Hybrid search plus text read from shared screensMeaning, exact terms and what was on the slidesSpeech that was never transcribed wellModerate to high
    Hybrid search, keyword and meaning together, is the combination that finds both "the Q3 renewal discount" and "the number we agreed for Acme".

    How do I find the exact moment in a long meeting?



    Index the transcript in short, overlapping passages of a few sentences, each carrying its start time, rather than one document per meeting, so a result takes you to the second the passage starts. Keep the meeting title, date and attendees on every passage so results can be filtered to a project, a customer or a date range before they are ranked.

    How do I make transcription good enough to search?



  7. Record each speaker on a clean channel where the tool allows it; crosstalk is the largest source
  8. of garbled text.
  9. Give the transcription service a vocabulary list of customer names, product names and jargon if it
  10. accepts one. Most modern services do.
  11. Search for names two ways: exact, and by meaning or fuzzy match, since a name can be transcribed
  12. several ways in the same meeting.
  13. Check a sample of ten meetings against what you remember before trusting the index.


  14. Recording and transcribing meetings has consent and retention rules that vary by country and state, so confirm them before you index calls with people outside your company.

    How do I do this with Mixpeek?



    Put the recordings in object storage you already use and connect the bucket. The multimodal extractor splits each recording into segments, transcribes the speech, embeds the transcript for search by meaning, and can read the text from shared screens in the frames, with every segment keeping its start and end time. A retriever then answers a question with the meeting and the moment, using keyword and semantic search together and filters on your own metadata such as customer or date. At the published rate of $0.05 per minute of video, indexing 100 hours of recorded meetings costs about $300 in processing.

    Related: why podcast search finds the episode but not the moment, why video search misses words that appear on screen, the best video transcription tools and the best speech-to-text APIs.

    Frequently Asked Questions



    Can I search all our meeting recordings at once?



    Yes, once their transcripts are in one index. Most meeting tools search one meeting's transcript at a time, so export the transcripts, or index the recordings directly, and search the combined index with the meeting title and date kept on every passage.

    Why doesn't searching the transcript find what I remember being said?



    Because exact search matches words, and people rarely remember the exact words. Search by meaning finds passages that express the same idea in different words. Transcription errors on names and numbers are the other common cause, which is why hybrid search helps.

    Can I jump to the moment in the recording, not just find the meeting?



    Yes, if the transcript is indexed in short passages that keep their start times. The search result then links to the second the passage begins instead of the start of the meeting.

    Can I search what was shown on screen during a meeting?



    Yes, by reading the text in the video frames with OCR and indexing it alongside the transcript. Slides, shared documents and dashboards often hold the figure people discussed without reading it out.
    Managed Mixpeek

    Put multimodal search to work

    Connect a bucket and Mixpeek runs the whole multimodal search pipeline for you: extraction, indexing, and search over your own objects. No models to wire up, nothing to host.

    Start with Managed
    MVS · bring your own

    Already have vectors?

    Keep your embeddings on your own cloud and run dense, sparse, and BM25 search directly on object storage. From $25/mo.

    Start with MVS

    Run this on your own data

    Point Mixpeek at the storage you already have and search your video, images, audio, and documents the way this guide describes. Build starts at $25/mo for up to 1M vectors.

    Search your own archiveRead Docs