Version: Latest

Audiocodes VoiceAI Connect

Use this channel to connect your Rasa assistant to Audiocodes VoiceAI connect.

Getting Credentials

To get credentials, create a bot on the VoiceAI connect portal.

  1. Select Bots in the left sidebar.
  2. Click on the + sign to create a new bot.
  3. Select Rasa Pro as the Bot Framework
  4. Set the bot URL and choose a token value.
  5. "Validate the bot configuration" once you have completed the "Setting credentials" section and the bot is running. Validation will only succeed if the bot is running and the token is correctly set.
Setting the bot URL with a tunneling solution when testing locally

Visit this section to learn how to generate the required bot URL when testing the channel on your local machine.

Setting credentials

The token value chosen above will be used in the credentials.yml:

audiocodes:
token: "token"

You can also specify optional parameters:

ParameterDefault valueDescription
tokenNo default valueThe token to authenticate calls between your Rasa assistant and VoiceAI connect
use_websockettrueIf true, Rasa will send messages through a web socket. If set to false, Rasa will send messages through http API calls.
keep_alive120In seconds. For each ongoing conversation, VoiceAI Connect will periodically verify the conversation is still active on the Rasa side.

Then restart your Rasa server to make the new channel endpoint available.

Usage

Receiving messages from a user

When a user speaks on the phone, VoiceAI Connect will send a text message (after it is processed by the speech-to-text engine) to your assistant like any other channel. This message will be interpreted by Rasa and handled by the dialogue engine, and the response will be sent back to VoiceAI Connect to be converted to a voice message and delivered to the user.

Sending messages to a user

Your bot will respond with text messages like with any other channel. The text-to-speech engine will convert the text and deliver it as a voice message to the user.

Here is an example:

utter_greet:
- text: "Hello! isn’t every life and every work beautiful?"
note

Only text messages are allowed. Images, attachments, and buttons cannot be used with a voice channel.

Handling conversation events

Non-voice events can also be handled by the bot. Here are a few examples:

EventintentDescription
startvaig_event_startVoiceAI will send this intent when it picks-up a phone call. In general, the response to that intent is a welcome or greeting message. Call context will be provided through entities
endvaig_event_endVoiceAI will send this intent when a call ends. You can use that to call an action that updates the call information.
DTMFvaig_event_DTMFVoiceAI will send this intent when receiving a DTMF tone (i.e user presses digit on the keyboard of the phone). The digit(s) sent will be passed in the value entity

The general pattern is that for every event sent, the bot will receive the vaig_event_<event> intent, with context information in entities.

Here is a simple rule to send a greeting message when a call to the bot is initiated:

- rule: New call
steps:
- intent: vaig_event_start
- action: utter_greet

Check the VoiceAI Connect documentation for an exhaustive list of events.

Configuring calls

You can send events from Rasa to VoiceAI Connect to make changes to the current call configuration. For example, you might want to receive a notification when the user stays silent for more than 5 seconds, or you might need to customize how DTMF digits are sent by VoiceAI Connect.

Call configuration events are sent with custom messages and are specific to the current conversation (sometimes to a message). Which means they must be part of your stories or rules so the same behaviour is applied to all conversations.

Those Rasa responses don't utter anything, they just configure the voice gateway. It is a good practice naming them differently, for example prefixing them with utter_config_<what_it_does>

All the supported events are exhaustively documented in the VoiceAI Connect documentation. We will look at one example here to illustrate the use of custom messages and events.

Example: changing a pin code

In this example we create a flow to allow a user to change a pin code.

- rule: Set pin code
steps:
# User says "I want to change my pin code"
- intent: set_pin_code
# Send the noUserInput configuration event
- action: utter_config_no_user_input
# Send the DTMF format configuration event
- action: utter_config_dtmf_pin_code
# A standard Rasa form to collect the pin code from the user
- action: pin_code_form
- ...

In the domain, we can add the utter_config_<config_event> responses:

noUserInput event

utter_config_no_user_input:
- custom:
type: event
name: config
sessionParams:
# If user stays silent for 5 seconds or more, the notification will be sent
userNoInputTimeoutMS: 5000
# If you want to allow for more than one notification during a call
userNoInputRetries: 2
# Enable the noUserInput notification
userNoInputSendEvent: true

DTMF event

utter_config_dtmf_pin_code:
- custom:
type: event
name: config
sessionParams:
# Enable grouped collection (i.e will send all digits in a single payload)
dtmfCollect: true
# If more than 5 secs have passed since a digit was pressed,
# the input is considered completed and will be sent to the bot
dtmfCollectInterDigitTimeoutMS: 5000
# If 6 digits are collected, VoiceAI will send those 6 digits
# even if the user keeps pressing buttons
dtmfCollectMaxDigits: 6
# If the user presses '#' the input is considered complete
dtmfCollectSubmitDigit: "#"

Now you can configure the pin_code slot in the pin_code_form to extract the pin code from the value entity with the vaig_event_DTMF intent:

pin_code:
type: text
influence_conversation: false
mappings:
- type: from_entity
entity: value
intent: vaig_event_DTMF
not_intent: vaig_event_noUserInput
conditions:
- active_loop: pin_code_form
requested_slot: pin_code

Notice how vaig_event_noUserInput was declared in the not_intent field.

Since the vaig_event_noUserInput intent is sent by VoiceAI Connect when the user stays silent as per our configuration, we must deactivate the form so we can pick up the conversation from a rule or a story and gracefully handle the failure.

In the following example, we simply cancel the current flow if we receive the vaig_event_noUserInput intent (i.e. user stays silent) while the pin_code_form loop is active.

- rule: Set pin code - happy path
steps:
- intent: set_pin_code
- action: utter_config_no_user_input
- action: utter_config_dtmf_pin_code
- action: pin_code_form
- active_loop: pin_code_form
- active_loop: null
- slot_was_set:
- requested_slot: null
- action: utter_pin_code_changed
- action: action_pin_code_cleanup
- rule: Set pin code - no response - cancel.
condition:
- active_loop: pin_code_form
steps:
- intent: vaig_event_noUserInput
- action: utter_cancel_set_pin_code
- action: action_deactivate_loop
- active_loop: null