node-red-doorbell

This service continuously listens for our doorbell ring sound. If it hears a ring then it will post a message on the specified slack channel.

Prerequisites

Only one docker container can access the microphone at a time. So you can not have 2 docker containers accessing the microphone !!!

Dockerfile

Following has been added to my docker file to enable audio utilities.

FROM nodered/node-red:1.2.6-12-minimal
USER root
...
RUN set -ex && apk --no-cache add sudo sox alsa-utils
# the following 2 command assure that node-red user belongs to the "host" audio group (gid = 63) of nuc-jan.
# this is needed to run commands like arecord, sox, aplay, alsamixer.
RUN addgroup -g 63 audio_host
RUN addgroup node-red audio_host

USER node-red
...

Docker compose file

The devices /dev/snd:/dev/snd are set in my docker-compose.yml file so that my doorbell container can access the microphone of my intel-nuc device.

  doorbell:
    image: janvda/doorbell:0.1.0
    ports:
      - "1889:1880"
    restart: always
    devices:
      # is needed to access host microphone
      - /dev/snd:/dev/snd

Environment Variables

It requires that following environment variables are set:

environment variable	description
slack_token	token of your slack workspace. It is something like xoxb-260...AlnL0Q
slack_channel_ring	ID of the channel where the ring event must be posted. The ID is typically a string that looks like C7MHSP2M7.

The flows also contain logic to upload recorded audio fragments to the set of training data or testing data of your edge impulse dashboard project. If you want to make use of this optional feature then the below environment variables must be set.

environment variable	description
EDGE_IMPULSE_APIKEY	api key for your project. You can find this in your edge impulse dashboard.
DEVICE_NAME	Unique identifier for your device - e.g. the MAC address.
DEVICE_TYPE	type of device - e.g the model.

Deploying new version of edge impulse

1. Go to Edge Impulse dashboard > Deployment menu

   • under Create library : choose WebAssembly
   • choose optimization int8 or float32
   • select build => this will download your model as zip file.

2. Goto the downloads folder where the zip file is located (see step 1). Unzip it and open terminal window within it.

3. Set DOCKER_HOST variable for terminal session so that it points to docker environment by using following alias

lan_setup

4. Copy the files to the respective folder through following 2 commands:

docker cp edge-impulse-standalone.js   nuc-jan_node-red2_1:/data/projects/node-red-doorbell
docker cp edge-impulse-standalone.wasm nuc-jan_node-red2_1:/data/projects/node-red-doorbell

5. You need to restart the node-red container (in portainer) to be sure it is using the new edge impulse

docker restart nuc-jan_node-red2_1

Publishing a new image version to docker hub

1. On your docker build machine (macbook in my case) you must have cloned the repository janvda/node-red-doorbell and assure that it has the latest changes.

2. Before doing this you can update the version in package.json and use this version as tag (see point 3.)

3. open a terminal window in the folder containing the Dockerfile of the cloned repository (see step 1) and execute following commands:

# 1. assure that DOCKER_HOST is not pointing to remote docker environment
export DOCKER_HOST=

# 2. build the image
docker build -t janvda/doorbell:latest .

# 3. publishing the image
docker login
docker images

# replace 7c2 in next command by the first characters of the IMAGE ID you have built in step 2
# replace 0.1.1 in following command by new version number
docker tag 7c2 janvda/doorbell:0.1.1
docker push janvda/doorbell:0.1.1
docker tag janvda/doorbell:0.1.1 janvda/doorbell:latest
docker push janvda/doorbell:latest

Running the docker image locally

This image (see previous section) can be run locally (e.g. on my macbook) using the command:

docker run -p 1880:1880 janvda/doorbell

# below sets environment variables based on my .env.node-red file
docker run -p 1880:1880 --env-file /Users/jan/Documents/15_iot/balena/active/nuc/nuc-jan/.env.node-red janvda/doorbell

# below command also shares the sound device - but this command doesn't work on macos !
docker run -p 1880:1880 --env-file /Users/jan/Documents/15_iot/balena/active/nuc/nuc-jan/.env.node-red --device=/dev/snd:/dev/snd janvda/doorbell

The node-red editor of the doorbell application in that case can be accessed at following URL:

• http://127.0.0.1:1880

developing / upgrading node-red-contrib-edge-impulse on my intel nuc

The janvda/node-red-contrib-edge-impulse repository is checked out in folder /data of my node-red2 container on my intel nuc.

1. This /data folder is accessible via samba. So I have mounted this samba share on my macbook.
2. Using visual studio code on my macbook I can edit the mounted samba share (=/Volumes/node-red2/node-red-contrib-edge-impulse) and push the changes to the github repository.
3. Note that the changes made to the repository in step 2 are not automatically used by the node-red flows. For that it is necessary to reinstall the node and restart node-red by means of following commands on my macbook:

# run alias to set DOCKER_HOST variable
lan_setup

# start bash session in the node-red2 container
docker exec -it nuc-jan_node-red2_1 /bin/bash

# following commands are executed in the bash session in node-red2 container
cd /data
npm install ./node-red-contrib-edge-impulse/
exit

# following commands are again executed on macbook
docker restart nuc-jan_node-red2_1
docker logs    nuc-jan_node-red2_1  -f  --since 20m

References

• janvda/node-red-contrib-edge-impulse

Relevant Edge Impulse Forum topics

• Better ML models with the Spectrogram block => see especially my second response summarizing the first test results (2020-11-17)
• Strange MFE spectogram for 1000 Hz sine wave
• MFE spectogram is limited to High Frequency of 8000Hz
• Getting error "RuntimeError: abort(OOM). Build with -s ASSERTIONS=1 for more info." when using WebAssembly build
• Mapping "Mel-fiterbank" to audacity Mel spectogram
• Recognize my doorbell
• Mapping audacity “Mel” to MFCC parameters
• Deployed WebAssembly build gives different outcome compared to my edge impulse dashboard
• Audio Processing Block: Spectrogram and 2D Output
• Comparison MFCC versus MFE to recognize doorbell ring