30/01/2024 5:53 am
Topic starter
Notifications
Clear all
0
Can the Xylo be deployed to continuously sample audio, or is it always necessary to iterate between capturing a fixed number of audio events and then processing them, repeat. i.e. is there a capability to stream?
Also, once a graph has been trained and then is deployed to the device, are outputs only monitored using an IDE? How can the device forward output to a control system, like that of an autonomous vehicle?
1 Answer
0
30/01/2024 7:58 am
Xylo operate continuously in streaming mode. There is no audio buffering used, and audio inputs are processed immediately.
The Xylo Audio 2 HDK requires a USB host interface to a PC to operate. The Xylo chip itself uses an SPI digital serial bus for simple interfacing with an MCU or CPU. When you are ready to begin integration design then please contact us for details of direct interfacing.
rogene Topic starter
30/01/2024 9:02 am
@dylanmuir This is great news! I did not see a tutorial or anything in the API about audio streaming - I will check the latest on GitHub (vs the website). Thanks very much! I have another week or two exploring simulation mode and then I will be ready to circle back for more information on direct interfacing. Should I request that help here, or send you an email direct?
DylanMuir 30/01/2024 9:10 am
@rogene It depends on which mode you are using to interact with the HDK. Check this tutorial section : https://rockpool.ai/devices/quick-xylo/xylo-audio-2-intro.html#Part-III:-Deploying-the-AFE-and-SNN-cores-in-free-running-inference-mode
which describes operating Xylo Audio in a continuous inference mode using the "XyloMonitor" class.
