Amazon Web Services (AWS) released a guide on May 13, 2026, for developing real-time voice streaming applications using Amazon Nova 2 Sonic (Nova Sonic) and Amazon Kinesis Video Streams WebRTC. This architecture addresses latency, bandwidth constraints, and language barriers by replacing traditional WebSocket protocols with Web Real-Time Communication (WebRTC), a public protocol designed for peer-to-peer media streaming. The system utilizes WebRTC's native capabilities, such as adaptive bitrate streaming and forward error correction, to maintain audio quality across unstable network conditions.

The solution integrates Nova Sonic, which provides a unified speech-to-speech architecture for conversational AI, with Kinesis Video Streams. Users establish connections via signaling channels, transmitting audio and video data through bi-directional peer connections secured by Datagram Transport Layer Security (DTLS). The framework incorporates a server-side Voice Activity Detection (VAD) layer using a Gaussian Mixture Model (GMM) to reduce background noise and optimize speech accuracy before processing. Audio data is resampled to 16kHz and converted to Float32 format to meet the Nova Sonic API requirements.

The deployment guide provides architectural patterns for two real-world use cases. In a smart home scenario, the system uses an Amazon Bedrock Knowledge Base to retrieve data and interact with an MCP server to control IoT devices. In a connected vehicle application, the system monitors driver attentiveness through real-time audio and video feeds, establishing independent, encrypted WebRTC connections for safety supervision. AWS provides open-source samples and Python SDK integration on GitHub to assist developers in building responsive voice assistants for mobile and IoT devices.

Amazon Web Services (AWS) released a guide on May 13, 2026, for developing real-time voice streaming applications using Amazon Nova 2 Sonic (Nova Sonic) and Amazon Kinesis Video Streams WebRTC. This architecture addresses latency, bandwidth constraints, and language barriers by replacing traditional WebSocket protocols with Web Real-Time Communication (WebRTC), a public protocol designed for peer-to-peer media streaming. The system utilizes WebRTC's native capabilities, such as adaptive bitrate streaming and forward error correction, to maintain audio quality across unstable network conditions.

The solution integrates Nova Sonic, which provides a unified speech-to-speech architecture for conversational AI, with Kinesis Video Streams. Users establish connections via signaling channels, transmitting audio and video data through bi-directional peer connections secured by Datagram Transport Layer Security (DTLS). The framework incorporates a server-side Voice Activity Detection (VAD) layer using a Gaussian Mixture Model (GMM) to reduce background noise and optimize speech accuracy before processing. Audio data is resampled to 16kHz and converted to Float32 format to meet the Nova Sonic API requirements.

The deployment guide provides architectural patterns for two real-world use cases. In a smart home scenario, the system uses an Amazon Bedrock Knowledge Base to retrieve data and interact with an MCP server to control IoT devices. In a connected vehicle application, the system monitors driver attentiveness through real-time audio and video feeds, establishing independent, encrypted WebRTC connections for safety supervision. AWS provides open-source samples and Python SDK integration on GitHub to assist developers in building responsive voice assistants for mobile and IoT devices.