Tweaking Stream Video Transcripts

Stream and Public Access

Following my post about the Petri videos taped in Washington DC last week, a reader asked me why we didn’t post the videos on Microsoft Stream. The reason is simple: the Office 365 roadmap includes a work item (2778) to support Public anonymous external video sharing that will allow Stream videos to be marked for public access and embedded in a web site. Today, that’s not possible and viewers need Office 365 licenses to access Stream content.

Hopefully, Microsoft will deliver external access by the end of 2019 as the roadmap indicates. Apart from making Stream much more useful as a platform for corporate video communications (internal and external), it could help ease the issues guest accounts have viewing Stream content in Teams, Outlook Groups, and Planner.

Transcribing Human Conversations

The question prompted me to take a fresh look at Stream video processing, especially some of the advanced features like face recognition and automatic transcripts. I uploaded the 1.2 GB MP4 file for the conversation between Paul Thurrott and myself about Outlook, Teams, and the meaning of life.

If a video’s language is English or Spanish, Stream processes automatic captions after a video has completed uploading. The captions are then concatenated to form a transcript. The process happens faster than I remember and the transcript itself seems to be more accurate. In many cases, the problems visible in the transcript were due to the humans when we didn’t express thoughts clearly or mumbled. Obviously Microsoft has been working hard in this area to improve the quality of automatic transcription, which wasn’t great when they first launched Stream in 2017. Figure 1 shows the video playing. You can see the transcript in a scrolling pane on the right-hand side.

Talking about Office 365 for IT Pros with Paul Thurrott in Washington DC
Figure 1: Talking about Office 365 for IT Pros with Paul Thurrott in Washington DC

Editing a Stream Transcript

In most cases, the transcript serves as an adequate record of what was said, but if necessary the video owner (the person who uploads it to Stream) can edit the transcript to correct the text (Figure 2). Notice that the transcript is broken up into chunks of text that are tied to timestamps in the video (the captions). The transcript intermingles contributions from different speakers. For instance, the text I’m editing completes a remark by Paul with “Thank you” and then runs on to my brilliant response starting with “Yeah, well.” It would be nice if you could edit the transcript more comprehensively to separate out what each speaker says more obviously, but that might break the connection with the video timestamps.

 Correcting a Stream transcript
Figure 2: Correcting a Stream transcript

Searching Transcripts

Apart from making transcripts look better, another good reason for editing transcripts to correct errors and improve the clarity of communications is that Stream supports transcript search. Because it’s based on the captions used to create the transcripts, don’t expect a full-text search facility. Instead, by using simple search terms that are likely to be found in transcript chunks, you should get reasonable results. Figure 3 is an example where we search Stream for “Office 365.” Notice that Stream tells you where in the video the transcript identifies the search term. These are clickable links that open the video and take you to the place in the video where the search term was found.

Searching Stream transcripts
Figure 3: Searching Stream transcripts

Transition from Office 365 Video

Stream is obviously improving but the migration from the older Office 365 Video portal has been slow. The latest (this month) report from Microsoft says that we are now in Phase 1:

Phase 1 – Tenant Admin Opt-in (started rollout June 2019)
Open to customers in specific regions who are not using the Office 365 Video REST APIs, Office 365 integrations (Delve, SharePoint Home, SharePoint mobile, Enterprise Search), and are comfortable with other differences between Office 365 Video and Stream. See 
more details about the migration experience for this phase.

It would be nice to complete the process… except that my videos that are in the older video portal are now a tad old and some qualify for historic status!


For more information about Stream, read the Office 365 for IT Pros eBook. In fact, we’ve just rewritten all our Stream content and will bring Stream back into the main book for the 2020 edition.

6 Replies to “Tweaking Stream Video Transcripts”

  1. The transcription feature is ‘nice’ and the word recognition works better than I expected, but unless MS prefixes *who* is speaking, normal conversations become a garbled unintelligible mess. Would work OK if just one person is performing a long monologue, though.

Leave a Reply

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