tomcat-ASRAgent

Agent that listens to the microphone and outputs messages corresponding to real-time ASR transcriptions.

Example usage:

./asr_agent --mode websocket --mqtt_host "127.0.0.1" --mqtt_port 5556

To see all available options, run:

./asr_agent -h

Prerequisites

• C++ compiler that supports C++17 or higher.
• nlohmann::json library: https://github.com/nlohmann/json
• Boost Library: https://www.boost.org/
• gRPC++ and Protobufers: https://grpc.io/docs/languages/cpp/quickstart/
• Mosquitto: https://mosquitto.org/

Build process

mkdir build
cd build
cmake ..
make -j

Google Cloud engine

To use the Google Cloud Speech Recognition engine, you will need to point the environment variable GOOGLE_APPLICATION_CREDENTIALS to point to your Google Cloud credentials file.

If using the Google Cloud Speech Recognition engine, you can add word/phrase "hints" to tell the speech regognizer to favor specific words and phrases. These should be added to the conf/speech_context.txt file.

Building locally on MACOS

If using MACOS, there is an additional step required to configure an environmental variable for gRPC: ''' curl -Lo roots.pem https://pki.google.com/roots.pem export GRPC_DEFAULT_SSL_ROOTS_FILE_PATH="$PWD/roots.pem" '''

Docker instructions

Create the tomcat network:

The asr_agent uses an internal Docker network named tomcat to connect its components. It must be created before running docker-compose:

docker network create tomcat

Google Cloud Speech Engine:

docker-compose -f google.yml up --build

Vosk Speech Engine:

docker-compose -f vosk.yml up --build

By default, Docker will use connect to a mosquitto broker running on port 1883 of the host machine. If you are using a different Mosquitto broker, modify the mqtt_host and mqtt_port command line option in the asr_agent service entrypoint in google.yml or vosk.yml.