With apologies to Blake Shelton and Christina Aguilera fans, we’re not going to be discussing the latest results from the singing contest reality TV show. Instead we’re going to talk about the rise of speaking as a convenient and practical way for us to interact with our mobile electronics. Perhaps the best-known examples today are Siri (for Apple iOS-based devices) and Google Voice (for Android-based platforms).
First, let’s stop to consider why implementing voice recognition (something that comes naturally to almost all humans) is so difficult for mobile electronics. For one thing, the environment in which mobile devices operate is less forgiving than for non-mobile electronics. It can be noisy outside, and users are often moving relative to their devices. Those factors make it harder for the device to separate the voice signal from background noise. For another, voice processing takes a lot of computing power, which in turn can consumes a lot of power which is detrimental to battery life, a paramount concern for mobile device application developers.
That leads us back to Siri and Google Voice. These applications solve the computing power problem by putting the heavy-duty computing in the “cloud”. That approach can work fine if your device is always cloud-connected when you need it be. But it fails miserably if you have no connection or a poor quality connection. Also, the cloud does not provide a real-time response that consumers expect.
So what’s the solution? For many devices it is to move away from processor and cloud-based applications and towards embedded hardware-based applications. By implementing critical voice recognition building blocks in silicon, no round-trip to the cloud is required thus eliminating the slow response from cloud-based solutions while simultaneously reducing power consumption.
For example, QuickLogic’s EOS™ S3 Sensor Processing SoC platform (just recently production-qualified) includes a hardened subsystem specifically designed for always-listening voice applications. With its dedicated PDM-to-PCM conversion block, and Sensory™ Low Power Sound Detector (LPSD) technology, the integrated system supports a wide range of features which enable it to identify voice commands, even in extremely noisy environments. Typical power consumption is substantially less, far better than traditional MCU-based solutions, and no network connection is necessary.
Voice commands are an extremely convenient way for users to interact with their mobile devices, and now – thanks to hardware-based voice processing approaches – we have made it practical as well.