Offline Voice Control - DEEPCRAFT Voice Assistant - Demo 2

Background

Encouraged by the article "Offline Voice Control using DEEPCRAFT™ Voice Assistant" I was wondering how to make an own PCB - shrunk, compared to Infineon's official PSOC™ Edge E84 AI Evaluation Kit - so that it can easier be used for DIY projects, when e.g. voice control shall be added to any existing system or 3D printed projects.

Getting started

Well, first observation - all microcontrollers of the PSOC Edge series are available in BGA or WLB packages, what does not really make it convenient for self-made PCBs, especially when looking towards the assembly respectively soldering. However - with some curiousness on how to do it and having supporters like Aisler, as PCB manufacturer including assembly service, and Würth Elektronik, providing most passive and electromechanical components - let's start compiling a block diagram...

Block diagram

.... with the minimum what is required for running offline Voice Control:

• Microphone - to catch your voice
• Microcontroller - running DEEPCRAFT's AI models of the Voice Assistant
• I/O-interfaces - allowing connection to any target system - whatever shall be controlled by voice

Hardware / Schematic

The schematic gives an good overview of the system. Infineon's official evaluation kit's for PSOC Edge, PSOC™ Edge E84 AI Evaluation Kit and PSOC™ Edge E84 Evaluation Kit, have been used as a reference, but the microcontroller was changed to the less-featured PSOC Edge E83, as no graphical capabilities were mandatory - a I2C/SPI display can be connected when simple GUI is required.

The system is powered by using an USB type-C connector or a 2-pin header, and two linear voltage regulators (TLS208D1EJV33, TLS208D1EJV) are used to generate the required on-board voltages (3, 3V, 1.8V). Reverse voltage protection is done by BAS52-02V diodes. The external 128 MBit memory S25FS128SAGMFB100 connected by Quad-SPI keeps the AI-models, to process the data coming from the digital IM73D122 microphone.

Note: When using the MiniProg4 Program and Debug Kit (CY8CKIT-005-A) ensure to set the target power to 1.8V! Neither voltage regulation nor protection is done on the PCB when connected to J2 (SWD/JTAG header).

All collateral components as capacitors and connectors have been provided by Würth Elektronik.

PCB

Out from the schematic a PCB has designed using KiCad that has been produced

and assembled by Aisler. (PCB code)

Demonstrator

For better demonstration of the final project, beside of the optional 0.96" b/w OLED-Display a NeoPixel stripe using 10pcs WS2812b LEDs has been attached.

Software

The software project is based on the article "Offline Voice Control using DEEPCRAFT™ Voice Assistant" where you can find the details how to implement your own wake word and commands.

For this demonstrator the DEEPCRAFT™ Voice Assistant Web UI is used to define a wake-word (e.g.: 'Hey Aisler') and some commands (e.g.: "turn left") to control the LED strip.

After showing the start-up screen including some credential to the supporters, the system waits for the wake-word. If detected, the question "Command?" is shown on the OLED-screen, and the user may say within some period the command to control color and movement of the LED strip.

Implemented commands are:

• Change the LED color: Say "switch" or "lights" followed by a color or "on" / "off"defined colors are: "red", "green", "blue", "yellow", "pink" and "white"
• Animate rotation:Say "turn" followed by "left" or "right" to change the direction accordingly
• Stop animation: Say "stop" to light the LED strip steadily

Note: When using the MiniProg4 Program and Debug Kit (CY8CKIT-005-A) ensure to set the target power to 1.8V! Neither voltage regulation nor protection is done on the PCB when connected to J2 (SWD/JTAG header).

Note: As mentioned in the README.md in the project source files, either Arm Compiler for Embedded or LLVM_ARM need to be installed to compile it. GCC wont compile the ML firmware.

• Download LLVM compiler here: https://github.com/ARM-software/LLVM-embedded-toolchain-for-Arm/releases/tag/release-19.1.5 (file "LLVM-ET-Arm-19.1.5-Windows-x86_64.zip").
• Extract it into a directory on your computer. Do not use spaces or special characters within your path.
  e.g.: "C:/mycompilerdirectory/LLVM-ET-Arm-19.1.5-Windows-x86_64".
• Within the project edit common_app.mk and edit the CY_COMPILER_LLVM_ARM_DIR= to point to your install folder. Use normal slashes '/' instead of backward slashes '\'.
  e.g.: "CY_COMPILER_LLVM_ARM_DIR=C:/mycompilerdirectory/LLVM-ET-Arm-19.1.5-Windows-x86_64".

Mechanics / Housing

All together has finally packed into a 3D-printed housing, including battery and

simple-charger circuitry, so that it can be easily demonstrated every.

Video

The following video demonstrates the result

Summary

What DO YOU like to control by voice?

This project shows how to add offline voice control to any electronics project using the PSOC™ Edge microcontroller and DEEPCRAFT™ Voice Assistant turning spoken commands into actions your hardware can execute.

Admittedly, making and assembling a PCB for BGA package has some challenges,.I like to thank Aislerfor manufacturing the PCBs.I like to thank Würth Elektronik for providing the collateral parts.

Very Special thanks to Francisco and Theo for supporting the development and design of this project.

Well - that's it!... for the moment. Stay tuned!

Regards, Holger