Whether commuting to work or simply having fun around town, riding a bike can be a great way to get exercise while also enjoying the scenery. However, riding around on the road presents a danger as cars or other cyclists / pedestrians might not be paying attention while you try to turn. That is why Alvaro Gonzalez-Vila created VoiceTurn, a set of turn signals that are activated simply by saying which direction you are heading towards.

VoiceTurn works by using the Arduino Nano 33 BLE Sense at its heart to both listen for the “left” or “right” keywords and then activate the appropriate turn signal. Gonzalez-Vila took advantage of edge machine learning through the Edge Impulse Studio. First, he collected audio samples consisting of the words “left,” “right,” and then random noise via the Google Speech Commands Dataset. Next, he sent them through an MFCC block that does some processing to extract human speech features. And finally, the Keras neural network was trained on these features to produce a model. 

With the model deployed to the Nano 33 BLE Sense, Gonzalez-Vila developed a simple program that continually reads in a waveform from the microphone and passes it to the model for inference. Based on the result, a string of NeoPixels on either the left or right will begin to light up for a predetermined number of cycles. As seen in his video below, the VoiceTurn works really well at detecting keywords and is easy to see from a distance. You can read more about how this project was built in its write-up here.

Boards:Nano 33 BLE Sense

Categories:ArduinoMachine Learning

• 
• Tweet
• 

You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.