Hecttor logo
Hecttor SDK Docs
IntegrationsLiveKit

LiveKit Integration

The Hecttor LiveKit plugin brings real-time speech enhancement into a LiveKit Agents pipeline. Audio published by a participant is cleaned by the Hecttor SDK before it reaches your STT / voice AI stack, which improves transcription accuracy in noisy environments without you having to manage audio buffers yourself.

The plugin is available for Python and Node.js, distributed alongside your SDK license. Both languages release in lockstep under a single version.

When to use it

  • You already run a LiveKit Agents worker and want cleaner audio into your ASR / LLM pipeline.
  • You want the enhancement to run inside the agent process, with no extra network hop — all inference is local, and no audio leaves your machine.
  • You prefer a drop-in audio processing node over wiring the SDK into your own audio loop.

The plugin works with both LiveKit Cloud and self-hosted LiveKit servers: processing happens in the agent process, so the SFU deployment model does not matter.

If you are not on LiveKit, use the direct SDK integration instead — see Orpheus Getting Started.

How it works

The plugin implements LiveKit's FrameProcessor interface — the same noise-cancellation slot used by LiveKit's first-party plugins — and is passed to AudioInputOptions(noise_cancellation=...) (Python) or inputOptions.noiseCancellation (Node.js):

  1. The LiveKit agent subscribes to the remote participant's audio track.
  2. Each audio frame is passed to the Hecttor enhancer, which denoises it in real time. Frames are buffered internally into the model's chunks (20 ms by default); added latency is bounded by one chunk.
  3. The enhanced frames continue through your normal agent pipeline (STT, LLM, TTS).

By default the plugin wraps Orpheus (ASRSpeechEnhancer — models optimized for machine transcription, the right choice in front of STT), using a voice-isolation model that suppresses background speakers as well as noise. A perceptual mode wrapping Hermes (HumanSpeechEnhancer) is available for audio that people will listen to (call recording, listen-in).

Requirements

RequirementValue
LiveKit SDKlivekit ≥ 1.0 (Python) / @livekit/rtc-node ≥ 0.13 (Node.js); works with LiveKit Agents ≥ 1.0
Python3.9+
Node.js18+
Hecttor SDKInstalled alongside the plugin — see the version floor in Getting Started
PlatformsSame as the Hecttor SDK: Linux x64 / arm64 (glibc 2.28+), macOS, Windows x64. The plugin package itself is platform-independent.
Hecttor SDK keyRequired — issued during onboarding

Next steps

Continue to Getting Started for installation and the integration steps.