How Microsoft Search Finds Spoken Text in Teams Meeting Transcripts

Search for What Participants Say During Teams Meetings

Microsoft message center notification MC260749 (last updated August 12) titled Microsoft Search: Find a meeting recording based on what was said is both technically interesting and important. Described in Microsoft 365 roadmap item 82003, the roll-out was delayed several times, but the way is now clear for Office 365 tenants to be able to search videos using spoken text along with a bunch of other changes to make Teams meeting recordings more accessible and useful. While it’s hard to say exactly when individual tenants will have all the functionality described here, I expect worldwide deployment to be complete by the end of October 2021.

Everything in OneDrive

Exposing the content of meeting recordings for search is important because it starts the process to close a major compliance gap. Up to now, transcripts for online meetings have not been searchable. The problem first surfaced when Teams stored its recordings in Stream. When the meeting finished, Stream processed the recording and created the transcript. However, the transcript remained in the Stream Azure service and was inaccessible to Microsoft Search. If something can’t be indexed by Microsoft Search, its content cannot be found by a search.

Microsoft completed the migration the storage of Teams meeting recordings from Stream to OneDrive for Business or SharePoint Online (ODSP) on August 16, 2021. All new meeting recordings from that date are in ODSP with the migration of older content from Stream to ODSP happening later. Microsoft is busy building out the rest of the Stream 2.0 platform to handle videos which don’t come from Teams. For instance, they’ve released a preview of the new Stream browser interface which supports access to videos stored in both ODSP and the original Stream store.

The move to ODSP removed the ability to create and replay transcripts for meeting recordings which exists in Stream classic. Starting September 20, Microsoft plans to remove some of the automatically-generated transcripts from older videos in Stream classic to prepare for the migration to Stream 2.0.

To fill the functionality gap, Microsoft introduced a transcription capability for Teams meeting recordings (a recent update means that if you record a Teams meeting now, you generate a transcript automatically). However, the issue of searchability remained. Because ODSP stores the recording files, Search could index file metadata like the name of the recording, but that’s about all.

The gap in indexing and searchability is now closed. Teams stores the spoken text captured during a meeting (including speaker attribution so you know who said what) and meeting metadata in the Exchange Online mailbox of the meeting organizer. Capturing the spoken text in mailboxes allows Microsoft Search to index the data and therefore makes it possible for searches to find this information. And as we’ll see, ODSP also holds a copy of the transcript to allow the words in the transcript to connect with segments in a meeting recording.

Exchange Mailbox Storage for Transcript Information

Teams stores transcript information in a folder called ApplicationDataRoot/93c8660e-1330-4e40-8fda-fd27f9eafe10/MeetingTranscriptCollection in the non-IPM part of the mailbox. Hidden means that the folder isn’t available to users through clients like Outlook, but its contents are available to administrative interfaces like Microsoft Search and programs like MFCMAPI.

Transcripts are captured as mail items. Examining the captured items with MFCMAPI, it looks like two properties for are most interesting:

  • TranscriptJsonBlob: stores the spoken text captured during the meeting. In Figure 1, you can see some captured text, including the name of the speaker. When users view the transcript in Teams, the information is displayed in a nicer format. It’s also possible to download transcripts in VTT or Word (DOCX) format.
  • TranscriptMetadataJsonBlob: stores metadata about the call.

Spoken text from a Teams meeting transcript stored in Exchange Online
Figure 1: Spoken text from a Teams meeting transcript stored in Exchange Online

Linking Words to Videos

The original implementation for Teams meeting recordings stored in Stream classic supported transcription, including the ability to edit the transcript to correct obvious errors. To allow Microsoft Search to find the MP4 file for a meeting recording based on words spoken during a meeting, a background process copies the transcript data captured in Exchange Online and indexes it against the recording to match segments of the video with the spoken words.

Replication of transcript data from Exchange Online to ODSP can take anything from 15 minutes to a day after the meeting ends. Once the process completes, you can search for text spoken in meetings and find recordings using the transcript (Figure 2).

Microsoft Search finds spoken text in a Teams meeting recording
Figure 2: Microsoft Search finds spoken text in a Teams meeting recording

Transcript Playback

Matching words in the transcript with meeting recordings (and eventually, any video stored in ODSP) allows concurrent playback for the two elements. Microsoft 365 roadmap item 82057, rolling out in September 2021, delivers a transcript pane for video playback (Figure 3). No ability is yet available for a video owner to edit the transcript.

Viewing a transcript of a Teams meeting recording alongside the video
Figure 3: Viewing a transcript of a Teams meeting recording alongside the video

Curiously, closed captions are available for only 60 days from the date of recording. In addition, Microsoft says that “Closed captions aren’t fully supported” if you move or copy a recording from its original location. Presumably, this is because the move might affect the link to the transcript data.

Making Transcription Available to More Teams Users

The option to transcribe meetings used to be restricted to accounts with enterprise E3/E5 and Business Premium/Standard. In early July, Microsoft made live transcription available (MC260564) for other licenses, including the E1, F1, academic, and Business basic SKUs, noting that this step improves the accessibility of Teams and makes meetings more inclusive for those who are hard of hearing. Microsoft followed up with MC280258 (August 24), to announce support for transcripts and captions in 27 additional languages (Figure 4) to join the previous support for U.S. English.

The new languages supported by Teams meeting recordings
Figure 4: The new languages supported by Teams meeting recordings

Another Compliance Gap Nearly Closed

All the information shared during Teams meetings is gradually coming within the scope of compliance policies. eDiscovery can already find chat, presentations, and documents, and the advent of indexed speech means that spoken comments should soon come within the scope of eDiscovery searches. This hasn’t happened yet, probably because of the work needed to export transcripts and videos in eDiscovery cases, but I am sure this capability is high on Microsoft’s agenda.

Although the captured text is sometimes inaccurate, capturing any record of spoken comments is better than nothing. As time goes by, the artificial intelligence technology used to analyze speech to create the transcript will improve in terms of accuracy and ability to handle accents.

So much change, all the time. It’s a challenge to stay abreast of all the updates Microsoft makes across Office 365. Subscribe to the Office 365 for IT Pros eBook to receive monthly insights into what’s happening.

5 Replies to “How Microsoft Search Finds Spoken Text in Teams Meeting Transcripts”

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.