Sprinklr supports real-time Speech-to-Speech (S2S) models through the OpenAI Realtime API. This capability enables Voice Bots to process and respond to spoken input with lower latency and more natural conversations.

Instead of using separate speech recognition, language processing, and text-to-speech components, the Voice Bot communicates directly with the S2S model for end-to-end audio processing.

How Speech-to-Speech Processing Works

When S2S is enabled, the Voice Bot uses the OpenAI Realtime API to process customer speech and generate spoken responses directly. This architecture reduces processing delays and improves conversational flow.

Before you can use an S2S model, the model must be onboarded in the customer environment. Contact your Sprinklr representative for onboarding and configuration assistance.

Prerequisites

• An approved Speech-to-Speech model is onboarded in the environment.
• The required feature flag is enabled.

Enable a Speech-to-Speech Model

1. Enable the Feature

Submit a support request to enable the following feature:

GEN_AI_VOICEBOT_REALTIME_API_ENABLED

2. Enable Speech-to-Speech in the IVR Flow

1. Open the IVR flow.
2. Navigate to the IVR node configuration.
3. Turn on Enable Speech-to-Speech Model.

This setting activates real-time Speech-to-Speech processing for the selected Voice Bot.

3. Select a Voice-Enabled AI Agent

1. Open the Redirect to Voicebot node.
2. From the Voicebot list, select the Voice-enabled AI Agent application.

This ensures that calls are routed to the Speech-to-Speech-enabled Voice Bot for real-time audio processing.

Considerations

• Speech-to-Speech support is currently limited to a subset of Voice Bot capabilities.
• This feature is being released through a controlled rollout.
• Contact the Conversational AI team for onboarding and configuration support.