De-noise

De-noise improves audio quality by filtering out background noise before it reaches your pipeline. This creates clearer conversations, especially in noisy environments. Zero Runtime supports two approaches to denoising:

RNNoise: a deep learning plugin that runs inside Zero Runtime.
Inference providers: speech enhancement via the Zero Runtime Inference gateway (Sanas and AIcoustics).

RNNoise

The Zero Runtime Agents framework provides real-time audio denoising through the RNNoise plugin, which runs inside Zero Runtime. RNNoise uses deep learning to tell speech apart from noise. It:

Removes background sounds like keyboard typing, air conditioning, and ambient noise.
Improves speech clarity and speech-to-text accuracy.
Processes audio in real time with minimal latency.
Works with the Pipeline in both cascading and realtime modes.

Initialize RNNoise and pass it to your Pipeline as the denoise component:

from zrt import Pipeline
from zrt.plugins import RNNoise, DeepgramSTT, OpenAILLM, ElevenLabsTTS, SileroVAD

pipeline = Pipeline(
    stt=DeepgramSTT(api_key="your-deepgram-key"),
    llm=OpenAILLM(api_key="your-openai-key", model="gpt-5.4-nano"),
    tts=ElevenLabsTTS(api_key="your-elevenlabs-key", voice="your-voice-id"),
    vad=SileroVAD(),
    denoise=RNNoise()  # Enable noise removal
)

Inference Providers

The Zero Runtime Inference gateway exposes denoise providers as the SanasDenoise and AICousticsDenoise classes. Import the one you want and pass it to your Pipeline as the denoise component.

Sanas
AIcoustics

SanasDenoise integrates Sanas for real-time speech enhancement and noise suppression. No required parameters.

from zrt import Pipeline
from zrt.inference import SanasDenoise

pipeline = Pipeline(
    # ... other config
    denoise=SanasDenoise()
)

AICousticsDenoise provides live audio cleanup with on-the-fly speech enhancement and background noise reduction. No required parameters.

from zrt import Pipeline
from zrt.inference import AICousticsDenoise

pipeline = Pipeline(
    # ... other config
    denoise=AICousticsDenoise(model_id="quail-vf-2.2-l-16khz"),
)

Choosing a Denoise Provider

Provider	Type	Language Support	Extra Config
RNNoise	Runtime plugin	Language-agnostic	None
Sanas	Inference	Language-agnostic	None
AIcoustics	Inference	Language-agnostic	None

Use RNNoise when you want a low-latency solution that runs inside Zero Runtime with no external dependencies. Use Sanas or AIcoustics when you prefer to route denoising through the Zero Runtime Inference gateway alongside your other components.

Get started

Modalities

Configure Pipeline

Tools and Capabilities

Turn Detection and Interruptions

Telephony Agents

Context Management

Observability and Monitoring

RNNoise

Inference Providers

Choosing a Denoise Provider

What’s Next

Turn Detection

VAD and Interruptions

​RNNoise

​Inference Providers

​Choosing a Denoise Provider

​What’s Next

Turn Detection

VAD and Interruptions

RNNoise

Inference Providers

Choosing a Denoise Provider

What’s Next