![\"\"](\"https:\/\/blog.mi.hdm-stuttgart.de\/wp-content\/uploads\/2017\/08\/architecturesnowboy.png\")
<\/a><\/p>\n<\/p>\n
Hardware<\/h2>\n
Before we start, make sure you have connected a microphone and speakers to your device. A regular USB webcam will also do the trick.<\/p>\n
To check if the devices are actually connected we use the command:<\/p>\n
arecord -l\n<\/code><\/pre>\nto check the microphone and<\/p>\n
aplay -l\n<\/code><\/pre>\nfor the speakers.<\/p>\n
At this point, it is important that you remember the card number and device number of the microphone and speaker you have chosen to use.<\/p>\n
This gained numbers we now have to write in our configuration file under ~\/.asoundrc<\/em>:<\/p>\npcm.!default {\n type asym\n playback.pcm {\n type plug\n slave.pcm "hw:1,0"\n }\n capture.pcm {\n type plug\n slave.pcm "hw:0,0"\n }\n}\n<\/code><\/pre>\nWhere the two numbers after hw:<\/strong> are the card number followed by the device number and should be replaced by your own values<\/em><\/p>\nMore information about the ALSA sound configuration can be found here<\/a><\/p>\nNow you should be able to record sound with:<\/p>\n
rec test.wav\n<\/code><\/pre>\nand hear your own voice with:<\/p>\n
play test.wav\n<\/code><\/pre>\nGoogle Assistant<\/h2>\nGoogle API Registration<\/h3>\n
First, if you don’t have a Google account already, create one. After you are logged in to the google developer console, create a new project. Now we have to enable the Assistant API<\/a>. Next, to create our OAuth ClientID we have to go here<\/a>. Select “Other”<\/strong> and when you finished download the json file, you can transfer it to your Raspberry Pi with:<\/p>\nscp ~\/Downloads\/client_secret_client-id.json pi@raspberry-pi-ip-address:\/home\/pi\/\n<\/code><\/pre>\nInstall Google Assistant Library<\/h3>\n
To download and install the libarary in a python3 enviroment we use the folling script:<\/p>\n
sudo apt-get update\nsudo apt-get install python3-dev python3-venv \npython3 -m venv env\nenv\/bin\/python -m pip install --upgrade pip setuptools\nsource env\/bin\/activate\npython -m pip install --upgrade google-assistant-library\npython -m pip install --upgrade google-auth-oauthlib[tool]\npython -m pip install --upgrade google-assistant-sdk[samples]\n<\/code><\/pre>\nAuthentification with google cloud<\/h3>\n
Before we can use the assistant we first have to authorize our application. To do this we call:<\/p>\n
google-oauthlib-tool --client-secrets \/path\/to\/client_secret_client-id.json --scope https:\/\/www.googleapis.com\/auth\/assistant-sdk-prototype --save --headless\n<\/code><\/pre>\nWe will receive a code on a website that we have to enter in the terminal.<\/p>\n
Run the sample<\/h3>\n
Once you authorized your application, you can now run the sample from google.<\/p>\n
google-assistant-demo\n<\/code><\/pre>\nYou now have built a regular google home and can trigger your running assistant with “OK, Google”<\/em> or “Hey, Google”<\/em>.<\/p>\nCustom Trigger<\/h2>\n
Since we are building our Google Home on our own, we are open to customization. So we will create our own Trigger. While others use a button to trigger there Assistant, I still want to use voice. I just wasn’t so happy with the hotword, so I replaced it with something more sympathetic to me: “Samantha”<\/em>, but you can choose anything you like.<\/p>\nSnowboy<\/h3>\n
![\"\"](\"https:\/\/blog.mi.hdm-stuttgart.de\/wp-content\/uploads\/2017\/08\/snowboylogo-300x68.png\")
<\/a>
\nTo create our own personal trained model for hotword detection we use the Snowboy Engine<\/a><\/p>\nCreate Hotword<\/h3>\n
Once you are logged in you have the option to choose a existing hotword or create your own.
\nIn any case, you will receive the model as the file to download. After downloading you also need to transfer this file to your Raspberry Pi.<\/p>\n
Python3 Support<\/h3>\n
Unfortunately, Snowboy<\/a> just delivers executable binaries for Python 2. To work with our Python 3 Google Assistant Library we need to build the binary on our own. To do this we first clone the Snowboy GitHub repository:<\/p>\ngit clone https:\/\/github.com\/Kitt-AI\/snowboy.git\n<\/code><\/pre>\nFirst, we have to make the snowboydetect.py file with swig in the directory \/swig\/python3\/<\/em><\/p>\nmake\n<\/code><\/pre>\nYou need to have swig v3.0.10 or higher – if you are stuck read the comments of the Makefile<\/em><\/p>\nAfter successful build you should be able to run the Snowboy sample under examples\/Python3\/<\/em> with:
\npython demo.py your-trained-model.pmdl<\/code><\/p>\nTrigger Google Assistant with Snowboy<\/h3>\n
The Last step is to connect Snowboy with the Google Assistant. That means when Snowboy detects your hotword, it will activate the Google Assistant, so he starts listening immediately. You don’t need to say “OK, Google”<\/em> anymore.<\/p>\nFirst, we change the snowboy demo.py file to call the google library:<\/p>\n
import snowboydecoder\nimport sys\nimport signal\nfrom googlesamples.assistant.grpc.pushtotalk import main\ninterrupted = False\n\ndef signal_handler(signal, frame):\n global interrupted\n interrupted = True\n\ndef interrupt_callback():\n global interrupted\n return interrupted\n\nmodel = "your-trained-model.pmdl" \n\n# capture SIGINT signal, e.g., Ctrl+C\nsignal.signal(signal.SIGINT, signal_handler)\n\ndetector = snowboydecoder.HotwordDetector(model, sensitivity=0.5)\nprint('Listening... Press Ctrl+C to exit')\n\ndetector.start(detected_callback=main,\n interrupt_check=interrupt_callback,\n sleep_time=0.03)\n\ndetector.terminate()\n<\/code><\/pre>\nAfter that, we modify the google library file at the end in you python environment under \/env\/lib\/python3.5\/site-packages\/googlesamples\/assistant\/grpc\/pushtotalk.py<\/em> and replace following code at the end of the file<\/p>\nwait_for_user_trigger = not once\nwhile True:\n if wait_for_user_trigger:\n click.pause(info='Press Enter to send a new request...')\n continue_conversation = assistant.converse()\n<\/code><\/pre>\nwith<\/p>\n
wait_for_user_trigger = once\nwhile True:\n if wait_for_user_trigger:\n return\n continue_conversation = assistant.converse()\n<\/code><\/pre>\nNow you should be able to trigger you assistant with your own hotword by running the Snowboy demo.py<\/em>.<\/p>\nConclusion<\/h3>\n
I hope you got some inspiration for further experiments with the Google Assistant. Another interesting point is, for example, to jump straight into your own google voice app.
\nIf you want to know how to build a google voice application, check out my other post here<\/a><\/p>\nThanks for reading through my post. For feedback or questions feel free to contact me.<\/p>\n
Cheers!<\/h4>\nMoritz Gruber<\/h4>\n