Tips for improving the accuracy of automated captions

1. Don’t underestimate the importance of good quality audio  

It is essential that the audio in your media is good quality for the speech recognition technology to be able to recognise the words being said.  

Here are a few things you can do to improve the quality of your recording and therefore make the automated captions more accurate:  

• Try to record in a quiet room so that background noise is at a minimum.  
• The fan on your computer can cause a lot of noise on recordings, so close any applications that you’re not actually using to minimise any noise.  
• Use an external microphone and place it correctly so that your voice is recorded clearly. Consider the distance of the microphone from your mouth, the direction it’s pointing in, avoid rustle from fabric if it’s a lapel mic, and minimise tapping from your keyboard if you are using a desktop mic . Note that a headset microphone will usually be better than using the microphone on your computer. The exact placement will differ between people and type of microphone, so do two or three short test recordings to try it out. Even if you’re using your device’s in-built microphone, try to position your device so that your voice is being recorded well.  
• A range of good quality microphones to suit your needs, recording location and environment can be borrowed from the Learning Spaces Technology team;  you can collect equipment from the Main Library.  
• Listen back to your tests on some headphones or earphones. What sounds best? You’ll soon arrive at the optimal position.  
• Advice on choosing and using a microphone  
• If you are recording one of your lectures in an equipped teaching space, be sure to wear the lapel microphone provided in the room and make sure it’s clipped onto your clothing where it can pick up your voice easily.  If the audio quality is poor, please report it as a fault to the IS Helpline. 

2.  Consider your voice  

Speak directly into the microphone. Don’t speak too fast, just try to speak clearly and naturally, all good practice when speaking to groups of people (face-to-face or remotely). Again, testing in advance is helpful to see what gives the best results.  

3. Find out more by attending one of our training sessions  

The Digital Skills Team offers regular training which covers how to request and edit captions on Media Hopper Create, along with guidance on good practice and getting the clearest audio for more accurate automated captions. To date,  over 120 people have learnt to caption on these courses. You can find more information and book a place via People & Money Learning, or use the course link here:

'Subtitling for media creators' Seminar Booking  

We run training on Media Hopper Replay (lecture recording) which covers creating captions and editing using that service.   Book onto the Media Hopper Replay (advanced) training course. 

4. Check your internet speed  

As the quality of any recording made on a video conferencing platform also relies on your internet connection, making sure that it’s as robust as possible by using a wired connection.  If you are recording without a live audience, consider recording using an application that doesn’t rely on an internet connection like Media Hopper Create’s desktop recorder.  You can then upload the media after you’ve recorded it.  

5. More advanced options   

a. Video conferencing applications like Zoom and Teams have in-built features which aim to minimise background noise or echo when you’re running and recording meetings. While this helps in many meeting situations, it can also mean the audio recording of actual speech is poorer than it might have been if you and the other participants were in a quiet environment using good microphones (as may be the case, for example, in some research interviews). Adjusting these settings can mean that the recording is more faithful to actual speech and therefore the likelihood of getting better automatic captions when you upload your video to Media Hopper Create is greater.  

• If you are recording via Zoom you can enable the ‘Original Sound’ option.  
• In Teams there’s an option to switch off the in-built noise suppression.  

b. If you have a recording which still has quite noticeable background noise, you can use software like Audacity  to reduce it. You’ll have to split the audio from the video using a video editing program such as  VLC Media Player, or another video editing program, clean it up in Audacity, and then rejoin it to your video in VLC Media Player.  

If you want to learn more about Audacity, book onto our Audacity training.