Published December 31, 2022 © GPL3+

Grove - Offline Voice recognition sensor

Grove - Offline Voice recognition module is a cost-effective, accurate, and Low powered sensor that can recognize up to 150 voice commands!

IntermediateFull instructions provided1 hour1,218

Grove - Offline Voice recognition sensor

Things used in this project

Hardware components

Seeed Studio Grove Offline voice recognition Module

Story

Story:

Seeed Studio introduced "Grove Design Co-Design Campaign" in which they urged makers to create new sensors for their Grove category. I was approached by my friend Sharyn, from Seeed Studio and was invited to participate in this.

After discussing possible new sensors ideas, I created this design and sent it to Seeed Studio to manufacture. Each approved sensor design gets 10 PCBs in which two were fully assembled and shipped to me.

Introduction:

Online Voice recognition modules require reliable internet connection to generate response, are not independent from the Microcontroller/Processor controlling them. While one can create an offline voice recognition module using a compatible processor like RaspberryPi, train a custom neural network to do this, it is not feasible for battery operated edge devices.

AI-Thinker's VC module series is apt for low power pure-offline voice recognition, based on US516P6 chip launched by Unisound. It has up to 150 pre-programmed commands that can be used to control your smart appliances. It features an identification time less than 100ms, in a compact form factor of 18mm * 17mm. The module can communicate through peripherals like UART/I2C/PWM/SPI/GPIO, I chose VC02 module with UART as output in my design.

Sensor Design:

I use Altium Designer to create my PCBs. VC-02 Kit schematic is available on the docs website. I created my schematic by referring to it. Since it is a relatively new module, I had to create the Altium library of the module (Get it here)

4 pin connector for UART is the standard connector used in Grove sensor, and so are the speaker and Mic connectors. I have added testpoints for the JTAG connections of the module, to upgrade the stock factory firmware if needed.

PCB Layout with Grove design dimensions

Features of PCB:

Standard Grove outline, 40x20 mm footprint
Power LED for VCC indication
Wake-up button to manually wake up the module
3 LEDs to indicate, React up - Cool light - Warm light

PCB dimensions and outline adheres to Grove sensor family, Hence is directly deployable in your projects. It is manufactured in blue silkscreen with ENIG finish. Quality of PCB manufactured is exceptional. Seeed Studio Fusion also has options to test your PCBA projects at their end before shipping, which can save a lot of your testing time for mass production products. A basic voltage test can also be handy to detect faulty units right at the time of production.

Assembled PCBs

I ordered a few more VC-02 modules and soldered the rest of the PCBs myself, Which wasn't time consuming as the Sensor requires just a few components to work.

Soldering Video (1)

Results:

I tried quite a lot of voice commands given in the documentation, The module works reliably and can be used for automating you home. A good microphone paired with this module can get up commands from anywhere in the room, 150 commands per room is quite a lot.

RESULTS!

Not everyone will like the response messages and would like to have custom input commands, for which AI-Thinker has a firmware generation website, Which is free to register and use, But does not have enough documentation done on it.

I was able to create a firmware with custom commands, However the firmware generated by that website will overwrite the UART firmware upgrading protocol inside the module, Hence a dedicated JTAG programmer is required to redo the firmware (As stated by AI Thinker engineer). Since this module already has Testpoints for JTAG port, In future the same module can be used once AI Thinker releases more information on firmware part.

For now, the possible workaround is to have Arduino UNO read the HEX response through the UART connection, and use an MP3 module to respond with desired phrase. But that is for future, as will it be a good project start 2023 with :)

Code

Grove Offline Recognition Module - Serial Test code

/*

###########################################

Designed by Anirudha Gaikwad, @_ro.box_ 
https://www.instagram.com/_ro.box_/?hl=en

Grove Offline Voice Recognition Module designed by me communicates over UART,
This code prints the HEX values sent by module over UART, 
List of commands: https://docs.ai-thinker.com/_media/vc-02_v1.0.0_specification.pdf



###########################################



*/


#include <SoftwareSerial.h>

#define SOFTWARE_RX 5   //RX connection of Grove Offline Voice Recognition Module
#define SOFTWARE_TX 6   //TX connection of Grove Offline Voice Recognition Module

SoftwareSerial Grove_Sensor(SOFTWARE_RX, SOFTWARE_TX); // RX, TX

void setup() 
{
  // Open serial communications and wait for port to open:
  Serial.begin(9600);
  while (!Serial); // wait for serial port to connect. Needed for native USB port only , This port is for displaying data Grove Sensor sends

  Serial.println("USB Serial Port connection Established!");

  // Set the data rate for the SoftwareSerial port - Grove Offline Voice Recognition Module runs at 115200 baud rate
  Grove_Sensor.begin(115200);

}

void loop() 
{ 
  if (Grove_Sensor.available()) 
  {
    Serial.print(Grove_Sensor.read(),HEX);
  }
 delay(10);
}