RoadTest: Matrix VOICE Dev Board
Author: dhanaganesh
Creation date:
Evaluation Type: Development Boards & Tools
Did you receive all parts the manufacturer stated would be included in the package?: True
What other parts do you consider comparable to this product?:
What were the biggest problems encountered?:
Detailed Review:
Initial impression:
Thank you element14 for delivering the Matrix voice safely with a good bubble wrapped package.
My first impression was very good, I wanted to try a cool-stuff first, even before recording something, tried colorful everloop with matrix voice. I'm interested to create a baby cry detection with the matrix voice. before starting my baby cry project i wanted to try some interesting stuffs seen in internet that i listed below after the snapshots.
Tests performed :
1. Alexa with Matrix voice
2. ODAS direction of sound with Matrix voice
3. PocketSpinkz with Matrix voice
4. Baby cry detection with Matrix voice
Alexa with Matrix voice
This was a kind of dream for me, i never tried building my own alexa using Rpi and stuffs i can say Matrix Voice performed very well in distance also with some fan noise i tried 2 meters and Rpi able to capture the expected stream from the Matrix voice. I felt good that whatever comes in my mind i asked to Matrix voice Alexa and almost whole day it was up with desk fan support. Wake-word support was not working in the sdk it was working intermittently but its just a software fix, i tried manual wake with sdk, it able to identify perfectly and replaying me without a delay. i even tried with the long distance, yes it heard me and worked like a charm.
Video demo:
Matrix Voice Rpi3 Alexa - Newark Element14Roadtest
ODAS direction of sound with Matrix voice
ODAS stands for Open embeddeD Audition System. This is a library dedicated to perform sound source localization, tracking, separation and post-filtering. i had a try with ODAS it is useful to check 360 degree audio detection. the main thing about this library is it is totally free(open source).
I have installed necessary software's to run the ODAS with Matrix Voice. i checked in my room in 2m radius distance it can able to identify the direction of sound, Its more-than enough for a baby cry audio detection in the bed room, i checked it can able to detect further distance also. 18 LED's on the Matrix voice shows the direction of the sound for us. Check the short video demo below.
Video demo:
Matrix Voice Rpi3 ODAS DirectionOfArrival - Newark Element14Roadtest
PocketSpinkz with Matrix voice
I do like the offline recognition very much as compared to the online as very less dependency and its easy to use for the devices which is not connected to the internet. In this situation the main player here is PocketSpinkz(Research). and they do gave me a positive response with the try with Matrix voice. Have a look on the video below that i have tried with matrix voice.
Video demo:
Matrix Voice Rpi3 pocketsphinx - Newark Element14Roadtest
Baby cry detection
This is the thing was stick in my mind for a while, the main aim about this project is to detect the infant/baby cry and based on the positive detection, we can do some action like alarming the parents/care takers. It would be more useful for motherhood to take care of their other soul. Its possible that we can identify babies laugh/cool/happy/sad/ sounds using the ML (Machine Learning). For the time being now i have used the open source project which is available here
The raspberry pi with Matrix Voice setup was done and required software's are installed to run the baby cry detection. Earlier i have trained the module in the PC and copied the required output files to the raspberry pi for running the detection with comparison of generated output from PC. and most of the time i was seeing the failure in identifying the baby cry, but i know i used very less samples that causes the problem. i was thinking that mic was not able to capture but when i try manually it was giving good output even in distance, mostly i tried the mono for this project.
Video demo:
Matrix Voice Rpi BabyCryDetection - Newark Element14Roadtest
i have spend my good time on identifying a baby cry and, the project i chosen is giving me a 40% accuracy which is not the adequate but the samples i used to train the model is very less, so i got a very less accuracy. I need to use a large number of samples (i don't have for now) to get a good accuracy, I'm planing to work on this in future (new project - open source).
Notes for Matrix Voice with Rpi users:
1. Have a small fan/cooling unit near your Raspbery pi for cooling down the ARM chip which is dissipating heat like anything.
2. If you are creating 3d printed setup make sure you that you are giving space for a small fan
3. Place a vertical plastic hand support in the other end of Matrix voice(one end already connected with Rpi GPIO connecter) so it will prevent the damage of GPIO's of Rpi & matrix voice.
Finally,
Matrix Voice seems performing as expected !! Just Go for it !!
