Conversation Relay: Dynamic Configuration, Application Reconnect, Echo Filtering, and ElevenLabs v4 Support

Twilio Conversation Relay now includes several new capabilities in General Availability (GA) to help you build more resilient and dynamic Voice AI agents:

• Dynamic configuration during a conversation: You can now update most session parameters (such as interruptibility, DTMF detection, timeouts, and partial speech results) and language attributes (such as ASR/TTS providers, speech models, voices, and hints) over your active WebSocket connection without dropping the call. This release introduces new inbound config, hints, and language WebSocket messages, an optional lang tag for inline multi-lingual text replies, and a new configUpdated outbound event.

• WebSocket application reconnect: Prevent dropped calls during infrastructure outages (such as autoscaling or container redeployments). When you set applicationReconnect="true" in your <ConversationRelay> TwiML, Twilio buffers up to 500 events during an unexpected connection drop and automatically replays them when it successfully reconnects to your application.

• Agent echo prompt filtering: Prevent the AI agent from interrupting itself or generating an incorrect response when its synthesized speech bleeds into the caller's microphone. Turn on echo prompt filtering by setting echoPromptFiltering="true" in your TwiML or in a dynamic config message.

• ElevenLabs v4 Turbo TTS support: You can now use ElevenLabs v3 Conversational and v4 Turbo text-to-speech models. These new models use natural-language audio tags like [speaking fast] instead of XML-based SSML tags and legacy numerical speed multipliers. 

For more details and implementation instructions, see the following documentation:

• <ConversationRelay> TwiML reference

• Update configuration during a conversation

• Conversation Relay WebSocket messages reference

• Conversation Relay best practices