Corporate Blogs

You say ‘po-tay-to’ – next steps for voice interface

By Trevor Wood - Last updated: Wednesday, August 23, 2017

Aksent speech recognitionAfter seeing the first Iron Man film, many people probably wished they had access to an all-powerful suit that could fly. Not me though, what I wanted was Jarvis, the voice-activated personal assistant who would act on my every request and perhaps even quip back at me.

In 2008, when Tony Stark made his first cinematic appearance, voice interface technologies were in the doldrums. Dragon Dictate had shown people what could be possible, but there was no reliable, cheap and widely available system.

In the 10 years since, the arrival of Siri, Cortana, Alexa and Google Assistant have changed everything. These new natural language voice interfaces unlock a host of new possibilities from controlling your smart home to setting an alarm, all without the need to pick up a device.

It is well-known that performance of the voice recognition algorithms can be patchy for people with accents from outside of North America. However, at Cambridge Consultants we think that the individual differences between users’ speech patterns don’t have to be entirely a problem. By analysing a speaker’s voice you could understand their emotional state or make a good guess as to where they are from or their age. This information could be used to target more appropriate responses.

We are currently developing a concept demonstration to classify your accent and provide interpretation and feedback on the particular nuances of your local dialect. We might even be able to train you how to speak in a completely different accent.

But to train our accent classifier we need your help.

Ready to get started? Come right this way.

We need lots of recordings of different speakers with English accents from all over the UK and US. To gather this data we have created a website – http://www.myaksent.co.uk/ – where visitors are encouraged to read a short paragraph of text and answer a few questions about their dialect. All the data is stored anonymously and will be used solely to train our speech analysis demonstration.

We really hope you can help us out by visiting the website and reading the ‘Please call Stella’ paragraph into your phone or laptop microphone. And be sure to follow our progress online. We will keep you updated with any interesting findings and, of course, share our accent analysis tool when it is complete.

Cambridge Consultants is investing in explainable AI as part of our wider investment in machine learning because we’d rather the future look more like Jarvis and less like HAL.

Click here to help us train our system


Media


AuthorTrevor Wood


Subscribe to blog feeds:


Recent Posts


Posts by categories