VIDEO: Editing Stream Transcripts
Four minutes, login to Office365 required
As long as there have been instructional videos, there has been a need for video transcripts or subtitles. A transcript posted with a video helps viewers who are deaf or hard-of-hearing, and helps with searching video content in video delivery systems. Creating the transcripts was a labor-intensive task, prone to error.
Microsoft Stream, however, can create a video transcript automatically. Not only does the finished transcript scroll along with the video as it plays, but it is time-stamped, and can be used to jump to a particular part of the video.
How accurate are the Stream transcripts? Surprisingly good and getting better. As far back as 2016, Microsoft voice recognition achieved an error rate of 5.9 percent, which sounds high but is comparable to the same error rate for human-generated transcriptions. Just four years later, the trained WER (Word Error Rate) has fallen to just 1 percent, exceeding the accuracy of average human perception.
But one percent error rate is “with training;” in practice it is higher.
Luckily Stream transcripts are easy to edit. Taking a few minutes to edit a Stream transcript improves the viewer experience, helps ensure you said what you wanted to say, and prevents the occasional embarrassing transcription error.
NOTES:
- WER, or Word Error Rate, is important to transcriptions and basic instructions, but is only a first step in AI comprehension of spoken language.
- GlobalMe Language and Technology: Speech Recognition Technology Overview