Voice-First

Reduce always-listening system power by 10x with RAMP

In the highly contested market for voice-first devices, Aspinity offers a leg up on competitive design solutions. Offering the industry’s only “analyze-first” architecture for always-on voice, Aspinity’s RAMP technology platform intelligently partitions system power and data resources, resulting in smaller, lighter form-factor devices with far less power consumed.

From smart speakers, wearables/hearables, remote controls and other smart home products, more than one billion battery-operated, wireless, voice-first devices will come to market by 2021. The problem is that today’s voice-first devices use an inefficient digitize-first architecture that digitizes and analyzes all incoming sound data for the wake word, even if the sound only occasionally includes voice. Since it’s the analog-to-digital converter (ADC) and other digital processors that typically dominate the power consumption of these always-listening systems, this digitize-first methodology wastes significant power processing irrelevant non-voice data.  

Analyze-First: Revolutionary Power Efficiency for Voice-First Devices

Aspinity’s RAMP chip offers a superior alternative. Its “analyze-first” edge architecture digitizes and analyzes only actual voice data to detect the wake word. That’s a major difference from existing solutions, which sift through all sound, including the long periods of time when there is noise but no voice, as they analyze sound while listening for the wake word.

As an ultra-low-power analog signal processor, the RAMP chip identifies voice directly from the raw analog microphone data at the earliest point in the audio signal chain. Just as our brain selects only the important sounds to send deeper into the brain for processing, the RAMP chip keeps the ADC, DSP and other heavy power-consumers further down the audio chain in a low-power mode — unless voice has actually been detected.

Analyzing voice data at the edge has its advantages. Only data containing voice data is digitized and analyzed for the wake word and transmitted to the cloud for wake-word verification. This eliminates the processing of irrelevant data and reduces the power of voice-first systems by 10x. For the first time, using an analyze-first architecture, designers can meet their power requirements in always-listening devices without sacrificing accuracy or features.

RAMP Technology Satisfies Preroll Requirements


Many wake word engines (WWEs) require 0.5s preroll (recent audio data) for increased accuracy of wake word verification. Unlike other solutions for voice activity detection (VAD) at the microphone edge, the RAMP chip uses a patented Aspinity processing approach to compress 500ms of preroll data. This preroll can then be reconstructed and used for wake word verification once voice has been detected. Alternatively, the wake word engine can be trained to use the compressed preroll data directly.

 

 

Learn more

Contact us for more information and to discuss how RAMP can improve the power and data efficiency in your device