Voice control software optimised for Synopsys's ARC DSP processor cores

Sensory’s TrulyHandsfree voice control software is optimised for the Synopsys DesignWare ARC EM DSP Processor Family. The release claims in typical 28-nm ARC EM5D processor chip consumes less than one microwatt of power executing Truly Handsfree low-power sound detection software and less than 40 microwatts operating in speech recognition mode.

So if you are designing a voice control based “always-on” IoT and wearable devices system, whether you're using whatever the processor and whatever the speech recognition software, these power consumption levels can be a benchmark to improve further in power consumption for voice analysis. With voice controlled IOT systems demanding lesser and lesser power consumption, designing systems with optimised power consumption is becoming very important.

While power consumption is important, designing robust voice controlled technology which can filter out noise and false triggers is still the most important part of your design. The noise or unwanted conversations and sounds need to be effectively filtered just like the way we human beings respond to what we hear. Huge amount of digital signal processing and need to be smart enough to skip unwanted Processing.

Sensory says its TrulyHandsfree voice control technology consistently recognizes phrases even when embedded in sentences and surrounded by noise, operating reliably in close to 0 dB signal-to-noise ratio real-world environments.

TrulyHandsfree can deliver more than 95 percent accuracy without false fires even in high noise and speech, according to Sensory.

Synopsys is focusing more on processor IP which includes DesignWare ARC EM DSP Family of processors, competing with processor IP vendors such as ARM and CEVA. Another EDA tool vendor Cadence also has a DSP processor family called Tensilica. All this means there are plenty of opportunities for SOC designers in IOT space in selecting processor core for digital signal processing.

Tweet Follow @eeherald