Silicon Ear

September 9, 2016 | 15:00
Before one day getting rid of buttons, keyboards, mice, touch screens etc, voice commands still have some way to go. Apart from the actual recognition and interpretation of words, it starts with the standby and detection circuits which must be simultaneously sensitive, refined, reliable at both near and far distances, workable in noisy environments, fast and – a crucial detail for the Internet of things (IoT) – have low power consumption. To input commands to our machines, today we use buttons, keyboards, mice, touch screens etc. Tomorrow, we will use voice commands to complement or even replace those devices. Our microcontroller ciruits associated with standby and voice detection will thus have to be simultaneously sensitive, refined, reliable at both near and far distances, workable in noisy environments, fast, and not least, have low power consumption.

The power consumption of current solutions for voice detection by a digital signal processor associated with an A/D converter in permanent wake mode is prohibitive for IoT applications powered by batteries, and even for smartphones. The Grenoble company Dolphin Integrations, a specialist in delta sigma converters for audio applications, offers with its Whisper Trigger a rapid start up from listening mode - which only consumes 20 µA. This amounts to a 10-20% of the power consumption of comparable devices. Its consumption in full working mode is around 360 µA. Its latency time is negligible; it wakes up in 1 ms.

For text recognition and detection of keywords, the Micro Electro Mechanical System (MEMS) mMOD67.5-COQ-VD.01 from Dolphin Integration will be associated with a DSP and a circular buffer on which the conversion, decimation and filtering will be performed.

(Click on picture below for a full version if it is not fully visible.)


Loading comments...
related items