SaynaAI/sayna

Sayna is a unified Voice Layer for AI Agents with a seemless integration to an existing agentic frameworks

What it solves

Sayna provides a high-performance, unified server for real-time voice processing. It eliminates the need to integrate multiple separate APIs for Speech-to-Text (STT) and Text-to-Speech (TTS), offering a single interface to manage various providers while handling the complexities of bidirectional audio streaming and noise reduction.

How it works

Built in Rust, the server acts as a coordinator between clients and voice providers. It uses WebSockets for real-time, bidirectional audio streaming and REST APIs for simpler tasks. The system employs a pluggable architecture that allows it to switch between providers like Deepgram, ElevenLabs, Google Cloud, and Microsoft Azure. To improve audio quality and interaction, it integrates Silero-VAD for voice activity and turn detection, and DeepFilterNet for noise suppression.

Who it’s for

Developers building voice-enabled applications, AI agents, or real-time communication tools that require low-latency audio processing and the ability to switch between different STT/TTS providers.

Highlights

  • Unified API: A single interface for multiple STT and TTS providers.
  • Real-time Streaming: WebSocket-based bidirectional audio flow.
  • LiveKit Integration: Support for WebRTC audio streaming and room-based communication.
  • Advanced Audio Processing: Integrated noise filtering via DeepFilterNet and ML-based turn detection using Silero-VAD.
  • Flexible Auth: Supports customer-based authentication delegated to external services.

Related

  • Project
    codeforequity-at/botium-speech-processing

    A unified API for open-source and cloud-based Speech-To-Text and Text-To-Speech services, simplifying audio processing for chatbots and voice applications.

  • Project
    lucasjinreal/Kokoros

    A high-performance Rust implementation of the Kokoro TTS model providing fast text-to-speech synthesis via a CLI, Rust crate, and OpenAI-compatible API server.

  • Project
    KoljaB/RealtimeSTT

    A Python speech-to-text library that provides fast, real-time transcription with integrated voice activity detection and wake word support.

  • Project
    KoljaB/RealtimeTTS

    A Python text-to-speech library designed for low-latency audio generation from strings, generators, and LLM token streams, supporting a wide variety of local and cloud TTS engines.

  • Project
    altunenes/parakeet-rs

    parakeet‑rs is a Rust crate that runs NVIDIA Parakeet speech‑recognition models (CTC, multilingual TDT, streaming, diarization, etc.) via ONNX Runtime. It offers offline and real‑time transcription with timestamps, speaker attribution, and GPU/CPU execution provider selection. Models are downloaded separately from Hugging Face; the library itself is MIT/Apache‑2.0 licensed.