High-end cars of various makes are gaining ears and a voice, allowing the driver to give instructions to the radio, the satnav, an in-built mobile phone, or the ventilation system. All of which is great for keeping attention on the road, where it belongs, and away from fiddling with buttons while ploughing through the central reservation.
There is a tension between making these speech interaction systems simple, and making them functional. After all, nobody wants a conversation with their dashboard that goes like this:
You: “Put a call in to the office.”
Car: “I don’t understand the instruction.”
You: “Call the office.”
Car: “I don’t understand the instruction.”
You: “Call Bob Smith.”
Car: “There are two entries matching Bob Smith in your address book: Bob Smith Office; and Bob Smith Mobile.”
You: “Call Bob Smith Office.”
Car: “You are driving too fast to make a call.”
You: “Aaaaargh!!”
Various suppliers - VoiceBox Technologies for example - are working with the major auto-makers to bring smart interactive systems to market that will actually respond in a meaningful way, rather than making you want to kick the centre console.
It’s not as simple a problem as detecting a PC users’ voice. Firstly, the user’s tone of voice is likely to change with speed, due to increasing road-noise levels, which will create a two-fold problem of extracting the information from the background and then extracting the meaning. Also, drivers will often have the radio playing just when they want to give verbal instructions, which will also pollute the audio environment.
One of the commonest tricks employed to improve speech recognition in difficult circumstances is to adapt to the user’s voice over time. Again, in a car, this might create problems. Cars are often used by more than one driver within a family or within a firm. And cars are typically sold on every three years or so.
All of which means that an adaptive voice recognition system will either need the ability to recognise multiple speakers, or will need a big fat reset button.