Live A2A translation user guide

Live audio-to-audio (A2A) translation allows agents to communicate seamlessly with customers in their own language. Follow these steps to enable live A2A translation for your customer service setup.

Before you begin

Ensure you have access to the following before you start the set-up process:

  • Google Cloud console with appropriate permissions.
  • Google Cloud SDK (gcloud) installed and configured.
  • jq installed for JSON processing.
  • Allowlist for translation features on your project.

Step 1: Create a conversation profile and phone number

Follow this process to attach a phone number to a conversation profile without using API calls.

  1. Navigate to the Conversational Agents (Dialogflow CX) console.
  2. Select your project that is on the allowlist: YOUR_PROJECT_ID.
  3. Click Create agent > Build your own.
  4. Choose the Global (Global serving, data-at-rest in US) region.
  5. Choose Flow instead of Playbook, then click Create.
  6. Select Integration in the panel.
  7. Click Manage > Create to obtain a new phone number.
  8. In the Conversation Profile settings, fill out the Display Name. Make a note of this name for later use.
  9. Save the phone number.

Step 2: Create a generator

  1. Navigate to the Agent Assist console.
  2. Choose the same project: YOUR_PROJECT_ID.
  3. Go to Live Translation.
  4. Click Create and enter a name for the generator.
  5. Choose your two languages.
  6. Select Voice in the translation channel.
  7. Select True for Conference Mode.
  8. Click Save.

Step 3: Configure your conversation profile

Follow these instructions to create a conversation profile with a Google Telephony Platform (GTP) number attached to it. This will create a conversation profile with a virtual agent.

The following steps help you to configure the conversation profile you created in step 1.

  1. Navigate to Conversation profiles > Suggestion types.
  2. Select Live translation and the generator you created in step 2.
  3. Enable Advanced suggestion settings to use unredacted conversation data for suggestion.
  4. Enable Telephony settings to use the bidirectional streaming API.
  5. Disable the Virtual agent in Choose to use Dialogflow, unless you want to configure a custom voice translation announcement (as described in optional step 5). If you want an announcement, keep the Virtual agent enabled.
  6. In Voice Configuration > Speech to Text Configuration, set:
  7. Audio Encoding: Linear 16
  8. Sample Rate Hertz: 16000
  9. Click Save.

Step 4: Verify the translation

Call the phone number you created in step 1 to verify if the translation is working as expected.

Optional Step 5: Configure a voice translation announcement

By default, conference mode directly translates audio streams without playing a preliminary greeting. If you kept the virtual agent enabled in step 3, follow these steps to configure a welcome announcement that plays before the live audio translation stage.

  1. Open the Conversational Agents (Dialogflow CX) Console and select your agent.
  2. In the Build tab, select the Default Start Flow.
  3. Click the Start page.
  4. Click the Default Welcome Intent route in the Routes section.

Update the Fulfillment

  1. Navigate to the Fulfillment section of this route.
  2. In the Agent dialogue > Text response section, remove any existing generic greeting (for example: Hi, how can I help you?).
  3. Enter your custom voice translation announcement (for example: This is an AI-powered translation service.)
  4. On the same Fulfillment panel, click Add dialogue response and select Live agent handoff.
  5. In the Metadata/Payload field for the Live agent handoff, enter an empty JSON object:{}
  6. Click Save.

Verify the announcement

When you place a test call, you should hear the announcement first. The empty JSON payload then initiates an immediate handoff, which transitions the call to the AGENT_ASSIST_STAGE where A2A Live Translation begins.