Google Assistant SDK for devices - C++

Linux / OSX build status

Windows build status

Requirements

This project is officially supported on Ubuntu 14.04. Other Linux distributions may be able to run this sample.

Refer to the Assistant SDK documentation for more information.

Setup instructions

Clean Project

If you have not setup this project before, you can skip this step.

sudo apt-get purge libc-ares-dev  # https://github.com/grpc/grpc/pull/10706#issuecomment-302775038
sudo apt-get purge libprotobuf-dev libprotoc-dev
sudo rm -rf /usr/local/bin/grpc_* /usr/local/bin/protoc \
    /usr/local/include/google/protobuf/ /usr/local/include/grpc/ /usr/local/include/grpc++/ \
    /usr/local/lib/libproto* /usr/local/lib/libgpr* /usr/local/lib/libgrpc* \
    /usr/local/lib/pkgconfig/protobuf* /usr/local/lib/pkgconfig/grpc* \
    /usr/local/share/grpc/

Build Project

1. Clone this project

git clone https://github.com/googlesamples/assistant-sdk-cpp.git
cd assistant-sdk-cpp

2. Install dependencies

sudo apt-get install autoconf automake libtool build-essential curl unzip
sudo apt-get install libasound2-dev  # For ALSA sound output
sudo apt-get install libcurl4-openssl-dev # CURL development library

3. Build this project

mkdir build && cd build
cmake .. -DCMAKE_STAGING_PREFIX=`pwd`/dist/usr/local
make install -j

4. Get credentials file. It must be an end-user's credentials.

• Go to the Actions Console and register your device model, following these instructions
• Move it in this folder and rename it to client_secret.json
• run get_credentials.sh in this folder. It will create the file credentials.json.

5. Start one of the run_assistant samples:

export LD_LIBRARY_PATH=`pwd`/dist/usr/local/lib
./dist/usr/local/bin/run_assistant_file --input ../resources/weather_in_mountain_view.raw --output ./response.wav --credentials ../credentials.json
aplay ./response.wav --rate=16000 --format=S16_LE

On a Linux workstation, you can alternatively use ALSA for audio input:

export LD_LIBRARY_PATH=`pwd`/dist/usr/local/lib
./dist/usr/local/bin/run_assistant_audio --credentials ../credentials.json

You can use a text-based query instead of audio. This allows you to continually enter text queries to the Assistant.

export LD_LIBRARY_PATH=`pwd`/dist/usr/local/lib
./dist/usr/local/bin/run_assistant_text --credentials ../credentials.json

This takes input from cin, so you can send data to the program when it starts.

echo "what time is it" | ./dist/usr/local/bin/run_assistant_text --credentials ../credentials.json

To change the locale, include a locale parameter:

echo "Bonjour" | ./dist/usr/local/bin/run_assistant_text --credentials ../credentials.json --locale "fr-FR"

Default Assistant gRPC API endpoint is embeddedassistant.googleapis.com. If you want to test with a custom Assistant gRPC API endpoint, you can pass --api_endpoint CUSTOM_API_ENDPOINT.

Enabling screen output

To get a visual output from the Assistant, provide a command to be run alongside every step of the conversation. It will execute that command along along with a provided argument of a temporary HTML file.

echo "what time is it" | ./dist/usr/local/bin/run_assistant_text --credentials ../credentials.json --html_out google-chrome

After you enter text, it will run google-chrome /tmp/google-assistant-cpp-screen-out.html.

If you prefer a different program, use that argument instead.