Native SDK examples and the ai-coustics-maintained LiveKit plugins require an
AIC_SDK_LICENSE environment variable. The browser quickstart keeps the SDK key on a backend and gives the browser a short-lived token. Get your license key from the ai-coustics developer platform. LiveKit Cloud authentication applies only to the LiveKit-owned plugins.LiveKit
LiveKit quickstart
Add ai-coustics speech enhancement, VAD and audio-quality analysis to a LiveKit Python or Node.js voice agent.
Python
Basic usage
Download a model from the CDN, load it, configure a processor and enhance audio in a synchronous workflow.
Async usage
Asynchronous version of the basic example using
async/await for model downloading and audio processing.File enhancement
Enhance WAV files from disk, downmixing multi-channel input to mono, with progress reporting.
Voice activity detection
Run a dedicated VAD model, tune sensitivity and hold duration and read speech decisions from the VAD context.
Audio analysis
Score an audio file with the Tyto analysis model across five-second windows.
Benchmark
Benchmark real-time processing throughput by spawning parallel processing sessions.
Pipecat
The Pipecat quickstart explains how to continue from loopback to the conversational example below, including provider credentials, VAD ownership and recording behavior. Use the matching Pipecat 1.11.0 dependencies.Speech enhancement (AICFilter)
Voice agent that enhances incoming audio with the AICFilter.
Voice activity detection
VAD event check without STT, LLM or TTS services. This example also applies enhancement before VAD.
Node.js
Basic usage
Download a model, create a processor and enhance audio samples with the Node.js SDK.
Async enhancement
Enhance audio with
ProcessorAsync and process concurrent streams without blocking the event loop.File processing
Read a WAV file, enhance it with configurable model and enhancement level and write the result to disk.
Voice activity detection
Run a dedicated VAD model and read speech decisions from the VAD context.
Async voice activity detection
Run
VadAsync on a worker thread, including detection combined with enhancement.Audio analysis
Buffer audio with
Analyzer and score it with Tyto using blocking or async analysis.Rust
Basic usage
Download a model at runtime, configure a processor with enhancement parameters and process audio.
Voice activity detection
Run a dedicated VAD model and read speech decisions from the VAD context.
Audio analysis
Score an audio file with the Tyto analysis model.
Parallel async
Process multiple streams concurrently with the async API.
Benchmark
Multi-threaded benchmark that measures real-time processing throughput across parallel sessions.
Build-time model download
Download a model at build time and embed it directly into your binary using the
include_model! macro.C
Enhancement
Load a model from file, create a processor and enhance mono audio using the C API.
Voice activity detection
Create an
AicVad from a dedicated VAD model and read speech decisions, with full error checking.Audio analysis
Buffer audio through a collector and score it with the Tyto analysis model.
C++
Basic usage
Load a model, configure a processor and enhance mono audio using the C++ API.
Voice activity detection
Run a dedicated VAD model and read speech decisions from the VAD context.
Audio analysis
Buffer audio through a collector and score it with the Tyto analysis model.
WebAssembly
Browser quickstart
Load a model and enhance audio in the browser with the WASM SDK.