talk2pi

Talk to your Raspberry PI with ReSpeaker, Snowboy and Google Speech APIs

Notice

Some files in this repository have been sourced from others to provide a central place for every required for this project. Please see individual scripts for copyright information.

Requirements

Raspberry Pi 4 (Recommended, but a 3 will be fine)
ReSpeaker 4-Mic Array by Seeed
Google account

Set up OS and ReSpeaker

Start with Raspberry Pi OS Buster Lite https://raspberrypi.org/downloads installed

First Positions

Make sure your on the network and have updated to the latest versions of everything with:

sudo apt -y update && sudo apt -y full-upgrade

Pre-requisties

Install dependancies:

sudo apt install git python3-pip
pip3 install spidev

Install Seeed drivers for ReSpeaker

See https://github.com/respeaker/seeed-voicecard

cd
git clone https://github.com/respeaker/seeed-voicecard
cd seeed-voicecard
sudo ./install.sh
sudo reboot

Enable LEDs

Enabled SPI interface:

sudo raspi-config

Go to Interfacing Options > SPI and enable

Install the sample code and test pixels:

cd
git clone https://github.com/respeaker/mic_hat.git
cd mic_hat
python3 pixels.py

(Confession: These are the instructions but I'm yet to make them work!)

Test Audio

Check the ReSpeaker has been 'seen' by the OS:

arecord -l

You should see something like:

**** List of CAPTURE Hardware Devices ****
card 1: seeed4micvoicec [seeed-4mic-voicecard], device 0: bcm2835-i2s-ac10x-codec0 ac10x-codec.1-003b-0 [bcm2835-i2s-ac10x-codec0 ac10x-codec.1-003b-0]
  Subdevices: 1/1
  Subdevice #0: subdevice #0

You can try a test recording:

arecord -D plughw:CARD=seeed4micvoicec,DEV=0 test.wav

Say a few words, then use Ctrl+C to stop

Play it back (assuming you're using the Pi's audio jack):

aplay -D sysdefault:CARD=Headphones test.wav

Please note if you have other devices connected, you may have different identifiers. Use arecord -L and aplay -L to get the possible values for the -D (device) switch.

Install and configure Snowboy

Snowboy provides the 'hotword' support for triggering 'full' recognition.

Full instructions here: https://github.com/mrpjevans/talk2pi/blob/main/step2_snowboy.md

Set up your Google service

Google's APIs provide near real-time text recognition.

Full instructions here: https://github.com/mrpjevans/talk2pi/blob/main/step4_google.md

Set up talk2pi

talk2pi.py is a starting point for building your own voice-controlled home assistant. It's basically a mash-up of the Snowboy code and the Google streaming code. With it you can add hooks to run your own code when certain commands are received.

Start by installing these dependancies (one time):

pip3 install gTTS pygame
sudo apt install libsox-fmt-mp3 libsdl-1.2 libsdl-mixer1.2

To try it out:

export AUDIODEV=hw:0,0
export GOOGLE_APPLICATION_CREDENTIALS="/home/pi/[your file]/json"
python3 talk2pi.py /path/to/hotword-model-file.pmdl

You should now be able to say your hotword and the code then 'flips' into Google mode. After a spoken statement it will go back into hotword mode again.

Try saying "what time is it". If you've got something hooked up to the audio jack, you'll get a spoken response (using Google's tex-to-speech engine)

Have a look at skills/tell_time.py for how to build you own skills.

mrpjevans / talk2pi