Three Years Until In-Car Voice Recognition Really Rocks

Speech recognition and NLU  need a lot more horsepower to work well in the car....

Source:  google.com

Voice recognition at its best in a car today is about 90 percent accurate. The goal of the car companies and of speech recognition software supplier Nuance is to reach about 95 percent accuracy.

And Nuance estimates that is about 3 years away.

A key ingredient to boosting recognition power is processor speed. In today’s cars, an infotainment system is running at about 300 to 1,000 million instructions per second (MIPS), said Nuance Director of Automotive Solution Architecture Brian Radloff. When you get to about 10 to 15X that level you can really support true natural language recognition, he said.

If you follow Moore’s law, with processor speeds doubling every year, we get to true natural speech recognition in a few years.

Som car makers claim they recognize natural speech right now, but it’s likely only for certain functions like making phone calls or searching through music.

Full natural speech recognition eliminates the need for key commands like “Call Bill,” and lets you say, “Please get Bill on the phone.”

“I believe the biggest gains to be made are going to be in conversational speech and understanding the intent of what the users is trying to accomplish. We’re starting to see that in telephony in the mobile space,” said Radloff.

It is also important that car makers improve other parts of the infotainment system to get the best speech recognition experience.  The graphics on the screen should properly tie into the speech control. And the microphone plays a role in the system as well.

“The biggest improvement we’ll see in system level performance is when the car company really takes a holistic view of voice and focuses on the whole package: the graphics on the screen, the microphone. Also, how do these systems handle a  recognition failure? Does it help you along if it doesn’t recognize if I said, ‘I feel like rocking today, let’s hear the Stones.’ Or it can say, ‘I don’t understand or it can come back and say “Please say the name of the artist again.’”

Radloff added, “The reality is, there are systems that are very sophisticated today and for some users, they deliver a very good experience. The bulk of the focus over the next 5 years in the automotive space and in voice in general is going to be how do we take this experience, that is very good for a certain group, and make it very good for a large swath of the car buying public.”

via http://www.speechtechnologygroup.com/speech-blog - Speech recognition and NLU  need a lot more horsepower to work well in the car.... Source:   google.com Voice recognition at its best in a car today is about 90 percent accurate. The goal of the car companies and of speech recognition software supplier Nuance is to reach about 95 percent accuracy. A ...