BBC R&D's Speech-to-Text: Revolutionizing Newsroom Efficiency and Audience Engagement (Full Transcript)

Speaker 1: BBC R&D's speech-to-text work has been a real success story for us. It's a great example of how a technology that's reached a tipping point of accuracy and affordability can suddenly find its way into production workflows.

Speaker 2: So a speech recognition system turns audio into a transcript. So for example, audio from BBC TV programmes or news bulletins, we'll automatically transcribe that into text. And the way they work is by learning a model to make those predictions, and we train that model by providing lots and lots of examples of bits of speech and bits of text. We could then use this, inputting it into the Caldi open-source speech-to-text software and training a deep neural network to recognise the audio. The accuracy increased to a point where BBC News were interested in using it within the newsroom.

Speaker 3: BBC News Labs is here to drive innovation in news. We're here to bring new technologies, new ways of thinking into the news production stream, and also for the benefit of our audiences.

Speaker 4: So the first tool that we developed using R&D's speech-to-text technology was Window in the Newsroom. So this takes a feed of all of the content coming into the main news video production system, and it runs it through the speech-to-text system to get a transcript out, and then it lets users search that content based on the transcript. We could save them kind of hours on a daily basis.

Speaker 3: The huge opportunity that's been opened up by R&D with this speech-to-text technology is to allow journalists to be journalists, to get on with the creative stuff, with thinking of that difficult question, with writing that headline so it really conveys the absolute essence of the story. So having a text element to the news makes it much more efficient and much more flexible.

Speaker 5: Audiogram is this tool that takes fantastic audio that we broadcast on Radio 1 and 1Xtra and it pops it online for us. It enables us to tweet it out, enables us to put it on Facebook. For us, it's so important because Radio 1's audiences are young, so it's really crucial that we're on those platforms. Without Audiogram, there are so many moments on Radio 1 and 1Xtra that we just wouldn't be able to bring to our audiences in the way it's enabling us to.

Speaker 1: So what next? Well, our team are working on three really interesting work streams. The first is real-time transcription of TV and radio, which we think will be particularly useful to teams working in a live environment, especially in news and sport. The second is speaker identification, which allows users to search for not just what was said in a programme, but who said it. And finally, we're doing work on general improvements to accuracy using larger language models and more effective algorithms.