As voice recognition becomes a de facto control technique for many tasks, there’s a growing move to handle some of the processing via cloud computing. This shift comes as software providers strive to make voice recognition even easier for automakers to deploy.
Cloud computing will let automakers and service providers bring far more functions to vehicles. That will increase the need for voice-activated control as the industry attempts to reduce driver distraction. Software companies are responding with enhanced algorithms that improve quality and reduce misunderstanding.
QNX Software Systems recently unveiled its Aviage Acoustic Processing Suite 2.0, which upgrades software that QNX said is already used on over 100 vehicle platforms. The new version is designed to go to the cloud when it doesn’t recognize user commands stored in the infotainment system’s library.
“Speech systems in the car may process commands like play and pause, but when you’re accessing navigation, it’s not practical to have every city and street on board. It’s much easier to send the data to the Internet,” said Andrew Poliak, Director, Business Development for QNX.
The combination of voice recognition and cloud computing is gaining momentum throughout the industry. Earlier this year, Pioneer Corp. rolled out Zypr, a voice-powered web application programming interface. It has a centralized, cloud-based architecture that helps developers access maps and routing, local search, social networking, and other data created by many content providers.
Ford recently made cloud computing a central aspect of its Evos concept car. Sync will be used to help users access either home or office systems and also let them listen to the same music stream when they move from home to car.
When voice data is sent outside the vehicle, it’s even more important to ensure that background noises are eliminated so the system won’t have to ask drivers to repeat terms. This requirement is growing as systems move to more conversational styles instead of using specific commands.
“When you’re doing off-board speech recognition, you need to clean up the audio coming out of the microphone. The s and t sounds often get dropped, which can cause difficulty for the voice-recognition system,” Poliak said.
The voice software has also been revised so it’s easier to deploy. Poliak noted that Aviage is self tuning, so it’s fairly straightforward to move the software from a small compact to a large luxury vehicle cabin. That can save a lot of time during the critical period of final development after prototype vehicles are available.
When engineers tune the voice-recognition system for a vehicle, they often need a prototype vehicle for a week or more so they can alter the voice algorithms for optimal performance. Many other teams also want access to prototypes, so shortening time can bring sizable benefits.
Poliak said that the latest iteration of Aviage is much simpler to use than competing technologies. One big factor is that it’s compact, so designers won’t necessarily need to include a digital signal processor.
“Basically, the library is so small and efficient that you can run on the main processor, so there is no need to offload the functionality onto a DSP. Other hands-free systems require a DSP or dedicated hardware,” Poliak said.
However, the software’s build environment, test and validation, and benchmarks do provide DSP compatibility for systems that use DSP for other tasks. QNX has developed versions that run on several common processors.