An open source telephony gateway for Google Dialogflow. All you need is a SIP trunk pointed to a server configured with the required software, and you're good to go.
- Full dialogflow telephony integration
- Call transfer via either SIP REFER or INVITE (requires support from your SIP trunking provider).
- playback interruption / barge-in via configurable hotword or phrase
- no activity detection
- recording support
- support for ambient typing sound while long-running fulfillment activity is happening
You'll need a server outfitted with the following software:
- drachtio server
- Freeswitch 1.10.1 (custom build to integrate with drachtio and dialogflow, see below for details)
- dialogflow module for Freeswitch
Of course, you will also need a dialogflow account, and a google cloud account.
The suggested linux distribution to run on is Debian 9. The easiest way to build yourself a server with all this is to use ansible, and create a playbook that runs the following roles:
- drachtio server ansible role
- freeswitch ansible role, builds v1.10.1 Freeswitch with support for grpc and drachtio
- nodejs ansible role
Create yourself a playbook like the following:
---
- hosts: all
become: yes
vars:
drachtioBranch: v0.8.2
vars_prompt:
- name: "build_with_grpc"
prompt: "Include the grpc modules (mod_google_transcribe, mod_google_tts, mod_dialogflow)?"
private: no
default: false
- name: "cloud_provider"
prompt: "Cloud provider: aws, gcp, azure, digital_ocean"
private: no
default: none
roles:
- ansible-role-fsmrf
- ansible-role-nodejs
- ansible-role-drachtio
Run it using ansible-playbook, answering 'True' to the question about install grpc support.
To authenticate with dialogflow, you will need to log into Google Cloud Platform and generate a service account json key file. Make sure you enable the APIs needed for dialogflow. Download the json file and place it on your server somewhere.
Then, edit your /etc/systemd/system/freeswitch.service systemd file, creating an environment variable named GOOGLE_APPLICATION_CREDENTIALS that points to it; i.e. add a line like this:
Environment=“GOOGLE_APPLICATION_CREDENTIALS=/home/admin/<your-service-key>.json
After doing that, reload and restart the systemd service:
systemctl daemon-reload
systemctl restart freeswitch
In the dialogflow console, make you have done the following:
- on the Speech tab, enable "Enable Automatic Text to Speech"
- for Output Audio Encoding, select "16 bit linear PCM
- on the General tab, enable "Enable beta features and APIs".
The application configuration file can be found in config/local.json. It consists of the following sections:
"drachtio": {
"host": "127.0.0.1",
"port": 9022,
"secret": "cymru"
}
specifies the location of the drachtio server to connect to.
"freeswitch": {
"address": "127.0.0.1",
"port": 8021,
"secret": "ClueCon"
}
specifies the location of the freeswitch server to connect to.
"logging": {
"level": "debug"
}
specifies log levels: info or debug
"dialogflow": {
"project": "<dialogflow-project-id-goes-here>",
"lang": "en-US",
"events": {
"welcome": "welcome"
}
"hotword": "OK Google"
}
project - the dialogflow agent to execute lang - language dialect to use, events.welcome - optional, if provided an event to send with the initial dialogflow streaming intent request hostword - hotword or phrase to use to "barge in" (i.e. interrupt audio).
"callTransfer": {
"method": "REFER",
"domain": "foo.bar"
}
method - indicates method to use for call transfer: REFER or INVITE domain - optional, if provided specifies the domain name to put in the Refer-To and Referred-By headers
"typing-sound": "/tmp/typing-sound.mp3"
Optionally, indicates the path to a .wav or .mp3 sound file to play while a fullfilment action is occurring. The audio will begin playing when an end of utterance has been detected, and will stop when the subsequent audio prompt from dialogflow arrives.
Note: an example typing sound file can be found in sounds/keyboard-typing.wav of this project.