I’ve been spending some time on YouTube recently, as we’re hosting a number of videos on the platform for a campaign we’re running to recruit foster carers.
Obviously we wanted to make sure we added captions to the videos, and to save a bit of time I’ve been trying out the caption tools. Firstly I noticed that YouTube automatically ‘machine transcribed’ some of the videos. On inspection, though, the accuracy of these was way off – possibly made worse by the particular accents which the speakers had, to which I doubt YouTube’s voice recognition is very well tuned.
However, the transcript tool fared much better. This lets you upload a plain text file with the transcript of the video. This is far easier than creating captions, which require timecodes.
YouTube then processes the file in a matter of seconds and applies the text as captions to the video, making a best guess at the timing. With the 7 videos I had uploaded, the results were very good, with split second accuracy in many cases.
One thing that did sometimes throw the timing a bit was where I had skipped some of the dialogue or shortened it for brevity and ease of reading. Overall, though, it was certainly satisfactory and a massive time saver.
In conclusion, then, I certainly wouldn’t rely on the auto-captioning, but it’s fantastic that YouTube are offering this given that the majority of uploaders will never bother manually adding captions. The transcription tool is an excellent compromise delivering better than average results, as it certainly worthy of praise.