CueLight demo

CueLight demo is an always-listening turn detector running in a web browser to process audio locally on a device.

CueLight-Nano is an always-listening turn detector that operates entirely inside a web browser. The model uses 0.5M parameters and takes up under 3 MB of space, ensuring that all audio processing happens locally and no audio leaves the user device. It is designed for applications requiring local turn detection in conversational audio. Users can test challenging conversational scenarios directly on the demo page, such as pauses during spoken requests. The project also provides access to the underlying code, model weights, and benchmarks on GitHub.

CueLight demo at a glance

Best for: Software engineer
Runs on: Web
Pricing: Free
Released: October 2026
Category: AI audio tool
Website: sshah03.github.io

Key features

Always-listening turn detector
Runs entirely in the browser
Local audio processing
Model size under 3 MB
Uses 0.5M parameters

Use it for

Detect conversational turns locally in a browser
Test turn detection on difficult conversational pauses

Alternatives to CueLight demo

audio.cpp — Runs modern audio models locally with high-performance, native C++ execution.
sherpa-onnx — Runs complex audio AI tasks offline on local devices and embedded hardware.
Pipecat — Builds real-time voice and multimodal agents with frame-level control.
AssemblyAI — Industry-leading speech models deliver high accuracy and low latency for voice applications.
Auricle — Runs entirely in the browser via Rust and WebAssembly with low latency and high accuracy.
FluidAudio — Runs local audio AI models on Apple Neural Engine.
kokoro-onnx — Runs local TTS inference with efficient ONNX runtime performance.
Free ASR — Zero-dependency real-time streaming speech-to-text with no account login required.
runs in your browser — It runs locally in your web browser using WebGPU.
Speaches — It acts as an Ollama equivalent for running local TTS and STT models.
WeNet — Achieves state-of-the-art results on public speech datasets.
Speakrail — Runs fully local on a single RTX 4090 with low latency and real-time interruption handling.

All CueLight demo alternatives