Maix Speech Recognition Demo

Quick Speech Recognition demo using the Maixduino Speech Recognition library on MAIX ONE DOCK with LCD screen and Microphone Array.

Prerequisites

PlatformIO
Sipeed-Kendryte210 Platform
- toolchain-kendryte210: 8.2.0
- framework-maixduino: 0.3.9
- framework-kendryte-standalone-sdk: 0.5.4
- tool-kflash-kendryte210: 0.9.1
Sipeed MAIX ONE DOCK (M1 Dock)
LCD Screen
Sipeed MAIX R6+1 Microphone Array

Environment Setup

Install PlatformIO(and vscode).
Add a new Platform, and select 'Advanced Installation'

Use the Sipeed k210 Platform Repository
https://github.com/sipeed/platform-kendryte210.git#e6ff97955c4cb15f7fe39450cadb8f145caefbf1

Import the Speech Recognition Demo

Select 'Sipeed MAIX ONE DOCK' as the board
Check 'Use libraries installed by Arduino IDE'

Library Notes

The Maix Speech Recognition library is included in this project as I needed to make a change to get this demo running. I was unable to get the Microphone Array functioning with the hard-coded pins so I added some input parameters to set them in 'begin()'. In order to not conflict with library included with Maixduino v0.3.9, they are renamed as Maix_Speech_Recognition_demo.h and Maix_Speech_Recognition_demo.cpp.

Running the Demo

After the demo is built and uploaded to the device, it should be displaying "Waiting for Wake Up..." on the LCD. This is the 'Standby' loop and will proceed when it is awake.

One or Two claps: This will move it to the second loop
Moving the Microphone Array a short distance across a table/desk: Nothing
Clearing you Throat: Nothing

The second loop will have it displaying "Color Command..." on the LCD.

One or Two claps: This will move it back to the standby loop
Moving the Microphone Array a short distance across a table/desk: Nothing
Clearing your Throat: Nothing
Red:     Change LCD display RED
Green:   Change LCD display GREEN
Blue:    Change LCD display BLUE
Cyan:    Change LCD display CYAN
Magenta: Change LCD display MAGENTA
Yellow:  Change LCD display YELLOW

While running, you will definitely notice that it recognizes wrong sometimes; this is pretty normal. I recorded the models on a desk in front of me, so it was fairly close; you may have to adjust your speaking distance. For certain colors you may have to stress some of the vowels as well to get it to register correctly.

For Example, "RED" will likely not be recognized if you say it very quickly with a short 'E'
If you say "RED" with a long 'E' it will have an easier time recognizing it.

Other reasons it will recognize incorrectly it due to how the recognizer itself works. It will pick the model it thinks is 'closest' to whatever it hears, even if that word/phrase was not included as a model at all.

Adding Additional Models

Models were recorded using the example present in the Maix Speech Recognition library examples

Just use that example to record new models and add them like the four model headers included in this demo. The example will output a 'number of frames' and an 'array' to a Serial Monitor.

Each Model entry should look something like this:
    const uint16_t fram_num_NAME = NUMBER_OF_FRAMES_HERE;
    const int16_t NAME_0[vv_frm_max*mfcc_num] = {
	ARRAY_HERE
	};