kugelaudio

What it is and what it does

KugelAudio is the official Python client for the KugelAudio text-to-speech API, handling speech synthesis via WebSocket streaming. The core SDK requires only httpx and websockets, making it lightweight to install. It provides a simple interface to generate audio from text using specified model IDs and save the output to files.

An optional turn-detection extra adds local, CPU-based conversation endpoint detection using ONNX Runtime and PyTorch. This feature downloads a version-pinned model bundle from Hugging Face, verifies SHA-256 checksums, and runs inference locally without network calls after the initial model cache. It's designed for latency-sensitive voice applications—particularly LiveKit Agents integrations—where you need to detect when a user has finished speaking. The turn detector supports multiple languages and provides detailed decision reasoning, including incomplete-turn timeouts and barge-in handling for overlapping speech.

Use it for:

• Generate speech from text in a web or mobile backend, streaming audio over WebSocket to reduce latency.
• Build a conversational voice agent that detects when users finish speaking using local turn detection without API round-trips.
• Integrate KugelAudio TTS into a LiveKit Agents application with manual turn handling and configurable silence thresholds.
• Implement a voice interface where you need to balance synthesis quality with low-latency streaming and local endpoint detection.
• Run turn detection offline after caching the model, suitable for edge deployments or privacy-sensitive applications.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Official Python SDK for KugelAudio's text-to-speech API, with optional local CPU-based turn detection for conversational applications.

Yes, if you need a KugelAudio TTS client or local turn detection for conversational AI. The core SDK is lightweight with minimal dependencies and active maintenance. The optional turn-detection extra adds real value for voice applications but requires Python 3.11+, Hugging Face access, and roughly 1.55 GiB of model runtime. No known security vulnerabilities. MIT license imposes no restrictions.

Install

kugelaudio on PyPI

pip install kugelaudio

uv add kugelaudio

poetry add kugelaudio

Installing kugelaudio

Before you install

Low friction: pure Python wheel with only httpx and websockets as runtime dependencies. Actively maintained with a release within the past week. Optional turn-detection extra requires Python 3.11+ and downloads a pinned ONNX model bundle from Hugging Face.

License in practice

MIT license permits commercial and private use with minimal restrictions, making it straightforward to integrate into proprietary applications.

Quickstart

pip install kugelaudio

from kugelaudio import KugelAudio

client = KugelAudio(api_key="your_api_key")
audio = client.tts.generate(
    text="Hello, world!",
    model_id="kugel-1-turbo",
)
audio.save("output.wav")

Requires a valid KugelAudio API key. Turn detection extra requires Python 3.11 or newer and Hugging Face authentication for the private model repository.

Verify before relying

• Whether the turn-detection model bundle size and download time are acceptable for typical deployment workflows.
• Performance characteristics and latency of turn detection on various hardware configurations.
• Availability and stability of the Hugging Face model repository for production use.

Package facts

Evidence: kugelaudio-1.9.0-py3-none-any.whl

Keywords: audio, streaming, text-to-speech, tts, websocket

Tags

More Speech packages