Using Gen AI to add subtitles on BBC Sounds

August 29, 2024 — From Aniruddh Dimri, Head of product, BBC Sounds:

“A crucial part of the BBC’s mission is that everyone across the UK feels the BBC is for them. Sometimes that’s about the content reflecting our different backgrounds, interests, and identity but it’s also about ensuring everyone can access our content.

“As an example, BBC Sounds currently produces approximately 27,000 hours of content per month – but much of it can be difficult to access for the approximately 18 million people in the UK who are deaf, have hearing loss or tinnitus in the UK. We have been exploring ways to add subtitles so people can follow the audio with the help of text.

“Doing this manually would be time consuming and prohibitively expensive.

“However, as we pilot new technology and explore how we can work with and use Generative AI tools to benefit our audiences, we have been looking at whether AI can help us add high quality, accurate subtitles to our audio content.

“We have been using a speech-to-text AI tool called Whisper AI, to quickly generate a high-quality transcript of the audio. This is reviewed by a member of the editorial team, and edited where necessary. A final transcript can then be uploaded with the audio on BBC Sounds.

“Today we’re starting to make this trial public. For the next three months we’ll be running a limited trial in which we’ll add subtitles and transcripts to a select range of audio content for the first time on BBC Sounds. This trial begins today on the web and Android app versions of BBC Sounds, and will come to iOS in the coming weeks. The programmes included in the trial will be In Touch, Access All, Profile, Sporting Witness and Economics with Subtitles, and we’ll look to add more programmes across the course of the trial. After the three-month trial period we’ll review progress and determine the next steps.

“The way it works is that on the web version of BBC Sounds, when playing programmes in the trial users will now see a new “subtitles” button in the play bar on the right hand side next to the settings and speed buttons. This looks the same as our accessibility icon on BBC iPlayer. Pressing this icon will make subtitles appear on screen, showing the words being said as they’re spoken just as they would on TV or BBC iPlayer.”

“On the BBC Sounds app users will see the same icon, this time under the play button. Pressing this will open a timed, full transcript of the full episode – which highlights the words being said as they’re spoken, and dynamically follows the text through the episode – a bit like how it looks when you follow the lyrics in some music apps.

“As mentioned – after three months we’ll review the progress made in the trial, how well the tools are working and if they’re a cost effective way of making transcripts available in Sounds. After that review, we’ll determine whether or not to continue, and if successful whether to roll them out to more of our content on BBC Sounds, and potentially to expand to our archive as well.”