Noise Suppression

In this tutorial, you’ll learn how to enable noise suppression for the Voice API and TeXML calls. Noise suppression works for both AI-powered calls (like AI Assistants and Gather Using AI) and regular voice calls. While it improves audio quality across all call types by reducing background noise, the biggest value comes from enhanced AI performance—cleaner audio leads to more accurate speech recognition and better AI responses. This makes noise suppression especially valuable for AI use cases where audio quality directly impacts user experience.

Voice API

The noise suppression can be enabled for the Voice API calls in the following way:

Don’t forget to update YOUR_API_KEY here.

The only parameter required for the request is direction which can have one of the following options: inbound | outbound | both. direction is defined from Telnyx’s perspective — relative to the audio streams on Telnyx’s side of the call, not to your application — which is the opposite of what the names suggest. The remote party is whoever is on the other end of the call (the caller on an inbound call, the callee on an outbound call):

Value	Audio that is denoised	What it means
`inbound`	Audio that Telnyx plays to the remote party	Cleans the audio the remote party hears from you
`outbound`	Audio received from the remote party	Cleans the audio you receive from the remote party
`both`	Both audio streams	Cleans audio in both directions

For speech recognition, streaming, and AI use cases, use outbound (or both): the audio you transcribe — the voice of the party you are listening to — is the outbound direction. This is also what Telnyx AI Assistants use internally.

Please be aware that the charge is applied for each direction separately.

    curl --request POST \
    --url https://api.telnyx.com/v2/calls/${call_control_id}/actions/suppression_start \
    --header 'Accept: application/json' \
    --header 'Authorization: Bearer ***' \
    --header 'Content-Type: application/json' \
    --data '{
        "direction": "inbound"
    }'

The noise suppression can be stopped at any time in the following way:

    curl --request POST \
    --url https://api.telnyx.com/v2/calls/${call_control_id}/actions/suppression_stop \
    --header 'Accept: application/json' \
    --header 'Authorization: Bearer ***' \
    --header 'Content-Type: application/json' \
    --data '{}'

Supported engines

Telnyx offers four noise suppression engines, each optimized for different use cases:

Engine	Value	Description	Best for
Denoiser	`Denoiser`	Built-in, general-purpose noise reduction	Default option for most calls
DeepFilterNet	`DeepFilterNet`	Open-source, full-band 48 kHz processing	Telephony and WebRTC
Krisp	`Krisp`	Telephony noise suppression with speaker isolation; supports multiple sub-models via `noise_suppression_engine_config.model`	Telephony with speaker isolation
AiCoustics	`AiCoustics`	STT-optimized noise suppression	AI and speech recognition workloads

Choosing an engine

For standard telephony, use Denoiser (default) or Krisp for speaker isolation.
For WebRTC calls, use DeepFilterNet for full-band processing.
For AI-powered calls (AI Assistants, Gather Using AI), consider AiCoustics for the best speech recognition accuracy.

Set the engine using the noise_suppression_engine parameter:

    curl --request POST \
    --url https://api.telnyx.com/v2/calls/${call_control_id}/actions/suppression_start \
    --header 'Accept: application/json' \
    --header 'Authorization: Bearer ***' \
    --header 'Content-Type: application/json' \
    --data '{
        "direction": "outbound",
        "noise_suppression_engine": "AiCoustics"
    }'

Engine configuration

Some engines support additional tuning via noise_suppression_engine_config. Parameters are engine-specific and ignored by other engines.

Krisp models

The Krisp engine supports several sub-models. Select a model using noise_suppression_engine_config.model:

Model value	Best for
`krisp-viva-tel-v2.kef`	Telephony; the platform default for 8–16 kHz audio
`krisp-viva-tel-lite-v1.kef`	Lighter telephony model
`krisp-viva-pro-v1.kef`	Higher-bandwidth audio; the platform default for 16–32 kHz
`krisp-viva-ss-v1.kef`	Krisp `ss` model

You can also set the suppression intensity with suppression_level (0.0–100.0):

    curl --request POST \
    --url https://api.telnyx.com/v2/calls/${call_control_id}/actions/suppression_start \
    --header 'Accept: application/json' \
    --header 'Authorization: Bearer ***' \
    --header 'Content-Type: application/json' \
    --data '{
        "direction": "inbound",
        "noise_suppression_engine": "Krisp",
        "noise_suppression_engine_config": {
            "model": "krisp-viva-tel-v2.kef",
            "suppression_level": 80.0
        }
    }'

DeepFilterNet configuration

The DeepFilterNet engine supports two tuning parameters:

Parameter	Type	Default	Description
`attenuation_limit`	integer (0–100)	`100`	Maximum attenuation applied to noise
`mode`	`standard` \| `advanced`	—	Processing mode

AiCoustics configuration

The AiCoustics engine exposes enhancement and gain controls:

Parameter	Type	Range	Description
`enhancement_level`	number	0–1	Enhancement intensity
`voice_gain`	number	0.1–4	Voice gain multiplier

TeXML

In TeXML there is a dedicated verb for enabling the noise suppression on the call.

<Response>
    <Start>
        <Suppression direction="outbound" noiseSuppressionEngine="Krisp"/>
    </Start>
...
    <Stop>
        <Suppression/>
    </Stop>
</Response>

Voice API Fundamentals

Voice Features

AI Capabilities

Tutorials

Regional & Restrictions

API Reference

For AI Agents

Noise Suppression

Voice API

Supported engines

Choosing an engine

Engine configuration

Krisp models

DeepFilterNet configuration

AiCoustics configuration

TeXML

​Voice API

​Supported engines

​Choosing an engine

​Engine configuration

​Krisp models

​DeepFilterNet configuration

​AiCoustics configuration

​TeXML

Voice API

Supported engines

Choosing an engine

Engine configuration

Krisp models

DeepFilterNet configuration

AiCoustics configuration

TeXML