Voice Layer for AI Agents Built with Rust, Pluggable to All Agentic Frameworks (github.com)

A groundbreaking voice processing server named Sayna has been launched, built with Rust to offer high-performance, real-time Speech-to-Text (STT) and Text-to-Speech (TTS) capabilities through WebSocket and REST APIs. This unified voice API consolidates multiple STT/TTS providers, including Deepgram, Google Cloud, ElevenLabs, and Microsoft Azure, allowing flexibility and adaptability for various applications. Key features include advanced live audio streaming through LiveKit integration, voice activity detection (VAD) using Silero-VAD, and optional deep noise filtering with DeepFilterNet. The significance of Sayna lies in its pluggable architecture, which enables seamless integration with various AI frameworks and can address the growing demand for efficient, multi-provider voice processing in AI-driven applications. By supporting real-time audio interactions and advanced noise management, Sayna opens new horizons for developers in fields like virtual assistants, telecommunication, and augmented reality. This innovative tool enhances user experience and engagement through its versatile functionalities, marking an essential advancement for the AI/ML community in voice/audio processing.