ESP32-Korvo AI Development Board Leverages ESP-Skainet Voice Assistant

Last September, Espressif Systems unveiled ESP-Skainet voice assistant optimized for ESP8266 and ESP32 wireless SoC with support for WakeNet wake word engine and MultiNet speech commands recognition with the former requiring just 20KB  RAM for one word, and the latter supporting up to 100 offline commands as long as you had 4MP SPI flash or more.

At the time, it only supported the Chinese language and worked on the upcoming “LyraT-Mini audio board“, now available for $26.99 shipped but only including one microphone. Espressif Systems has now announced a better AI development board with ESP32-Korvo AI development board includes featuring a mainboard with ESP32 processor and an audio ADC, and a subboard equipped with a 3-mic array, RGB LEDs, and various buttons.

ESP32-Korvo specifications:

  • Mainboard
    • Wireless module – ESP32-WROVER-B with ESP32 dual-core Wi-Fi / BT processor, 128 Mbit SPI flash,  and 64 Mbit PSRAM
    • Storage – MicroSD card slot
    • Audio
      • ES8311 I2S audio codec
      • Four-channel ADC with three channels for the microphones and one for AEC (echo cancellation) function.
      • Audio amplifier for external speaker up to 3 Watts
      • 3.5mm audio jack headphones
      • Speaker connector
    • USB – 2x Micro USB port (one for power, one for USB-UART)
    • Misc – Power switch, reset button, boot button, FPC connector
      to connect the mainboard & subboard.
    • Power Supply
      • 5V via Micro USB port
      • Battery header
      • 5V to 3.3V regulator.
  • Subbboard
    • 3x analog microphone arrays (spacing = 65mm).
    • 12x addressable RGB LEDs (WS2812).
    • 6x user-definable function buttons

Besides the new hardware, ESP-Skainet voice assistant has also improved with Multinet supporting both Chinese and English language, Wake Word Engine, and WakeNet wake word can be customized beyond the default “Hi, Lexin” with some common wake words like “Alexa” or a paid customization service that takes two to three weeks and requires audio samples from at least 500 different persons.  ESP-Skainet also performs audio processing to improve audio quality include AEC (Acoustic Echo Cancellation), AGC (automatic gain control), NS (Noise Suppression), VAD (Voice Activity Detection) and MASE (Mic Array Speech Enhancement).

ESP32-Korvo Block Diagram

You’ll find documentation to get started on Github, and can purchase the ESP32-Korvo AI development board for $35 on Digikey.

Support CNX Software - Donate via PayPal or become a Patron on Patreon
Advertisements
Subscribe
Notify of
guest
2 Comments
oldest
newest most voted
Jon Smirl
3 months ago

Skainet does work, but the accuracy is insufficient to pass Alexa qualification. Espressif is aware of that and they are working on alternatives., Meanwhile if you want certified Alexa on ESP you have to use external wakeword chip from one of the vendors with Alexa certification. Alexa certification is quite hard, you are only allowed to make three errors in 24 hours of testing.

I have no confirmation from Espressif, but Tensilicia does offer an AI coprocessor that could be integrated into a future ESP chip.

Drone
Drone
3 months ago

“Skainet”… reminds me of “Skynet”. Actually, anyone who understands the World Domination goals of the Chinese Communist Party (CCP) under Xi Jinping would put this under the all-encompassing “Xi-Net” moniker.

Advertisements