Artificial intelligence and machine learning (AI/ML) continue to revolutionize voice recognition, and products like Amazon’s Alexa have pushed user expectations. The technology works well for devices that are plugged in and connected to the cloud, but things get a bit more challenging for battery-operated devices. This is where Ambiq’s Subthreshold Power Optimized Technology-based (SPOT) excels.
Ambiq’s Voice-on SPOT (VoS) Kit targets the always-on voice-recognition space (Fig. 1). A mikroBUS Click Shield board adapter plugs into the processor board with an Apollo3 Blue Plus Cortex-M40-based microcontroller. A Vesper MEMSensing digital microphone click board is included with two slots available for additional peripherals. The main board includes a Bluetooth Low Energy (BLE) radio.
The processor keeps energy consumption to a minimum, but it’s the software running on the system that makes a difference (Fig. 2). Ambiq has partnered with a host of companies to address the voice-processing chain, including DSP Concepts for beamforming and noise reduction and Sensory and Retune DSP for keyword detection and local command processing.
The software can handle near-field (under three feet) and far-field (up to nine feet) voice processing with an excellent false rejection rate and high response accuracy rate, even with noise and music in the background.
The VoS platform was tested in a far-field, always-on-voice (AoV) application as a BLE remote control. A pair of AAA batteries will run the remote continuously for 409 days.
Although the kit targets mobile voice-recognition/command applications, it can be used for other IoT applications that require low-power AI/ML tasks.