Live A2A translation user guide

Caution: This feature has limited access. Contact your Google Cloud representative to access private GA features.

Live audio-to-audio (A2A) translation allows agents to communicate seamlessly with customers in their own language. To enable live A2A translation for your customer service setup, follow these steps in the Agent Assist console.

Step 1: Create a generator

1. Navigate to Live translation and click Create.
2. Enter a Generator name and target languages for the agent and the customer.
3. Navigate to Conversation channel and choose voice.
4. Enable Conference mode.
5. Click Save.

Step 2: Create a conversation profile and link a telephone number

1. 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.
2. Go to Choose to use Dialogflow and disable the Virtual agent.
3. Go to Voice configurations > SIP configuration and disable Allow virtual agent interaction.

Note: Attaching a phone number to the conversation profile creates a virtual agent with a conversation profile. The virtual agent is not required for the A2A workflow, but this is still the best way to get a phone number and link it to your conversation profile.

Step 3: Configure the conversation profile

Follow these steps to configure the conversation profile created in step 2.

1. Navigate to Conversation profiles> Suggestion types.
2. Select Live translation and enter the generator name you created in step 1.
3. Enable Advanced suggestion settings to use unredacted conversation data for suggestion.
4. Enable Telephony settings to use the bidirectional streaming API.
5. Enter 16000 for Sample Rate Hertz.
6. Click Save.

Step 4: Initialize conference mode

The conference mode service operates with the following workflow to handle mixed audio streams in a multi-party call:

1. Add your dedicated GTP phone number as a third participant to integrate live audio translation into an active call. Use the standard three-way conference functionality available on common telephony and contact center platforms.
2. The GTP immediately receives a single, mixed audio stream upon connection and processes it through a bidiStreamingAnalyzeContent request.
3. Gemini model synthesizes audio translation based on input audio and enables transcription for both original audios.It translates the attributed audio and plays the translated audio back into the three-way conference for all participants.