Skip to main content

​
Overview

Voice agents are vulnerable to background noise, especially other voices that trigger false interruptions. When this happens, your bot stops mid-sentence unnecessarily, creating a poor user experience. Pipecat Cloud’s Krisp integration processes incoming audio to remove ambient noise and background voices before they reach your agent, resulting in cleaner audio input with fewer false interruptions and better transcription.
Krisp processes the audio before it reaches your agent’s speech recognition and VAD systems, helping to ensure that only the intended user’s voice triggers interruptions.

​
Enabling Krisp VIVA

Enabling Krisp VIVA with Pipecat Cloud involves the following steps:

​
1. Use the Pipecat Base Image

Krisp VIVA dependencies are pre-installed in the official Pipecat base image. You don’t need to add any Krisp VIVA-related packages to your requirements.txt file or pyproject.toml file.
Krisp VIVA support is available in the dailyco/pipecat-base image starting with version 0.1.8.

​
2. Add the KrispVivaFilter() to Your Transport

In your bot code, import and add the KrispVivaFilter() to your transport: 
from pipecat.audio.filters.krisp_viva_filter import KrispVivaFilter
from pipecat.audio.vad.silero import SileroVADAnalyzer
from pipecat.transports.daily.transport import DailyParams, DailyTransport

transport = DailyTransport(
    room_url,
    token,
    "Voice AI Bot",
    DailyParams(
        audio_in_enabled=True,
        audio_in_filter=KrispVivaFilter(),  # Add Krisp VIVA filter here
        audio_out_enabled=True,
        vad_analyzer=SileroVADAnalyzer(),
    ),
)

​
3. Enable Krisp VIVA in Your Deployment

You can enable Krisp VIVA when deploying your agent through:
  • Config File
  • CLI
  • REST API
  • Dashboard
In your pcc-deploy.toml:
agent_name = "voice-starter"
image = "your-repo/voice-starter:0.1"
image_credentials = "my-dockerhub-secret"
secret_set = "voice-starter-secrets"

[krisp_viva]
audio_filter = "tel"

[scaling]
min_agents = 0
Supported values are:
  • tel: Supports up to 16kHz audio
  • pro: Supports up to 32kHz audio
By default, Pipecat processes input audio at 16kHz, making the tel value appropriate for most use cases.

​
Testing Your Integration

To verify Krisp is working:
  1. Deploy your agent with Krisp enabled
  2. Connect to your agent in an environment with background noise; try turning on a TV or podcast
  3. Observe whether background noises and voices affect your agent’s behavior

​
Pricing

Krisp processing is billed based on active session minutes:
Usage TierPrice
First 10,000 minutes per monthFree
Additional minutes$0.0015 per minute
For agents deployed with Krisp enabled, active session minutes count as Krisp usage, accruing towards your monthly bill.

​
Local Development

Krisp cannot be used in local development environments, as it requires a proprietary SDK and model which can only be distributed on Pipecat Cloud. When running locally be sure to not import the KrispFilter module and don’t apply it to your transport.

​
Troubleshooting

If you’re experiencing issues with Krisp noise cancellation:
  • Verify Krisp VIVA is enabled in your deployment configuration
  • Confirm you’ve added the KrispVivaFilter() to your transport configuration
  • Check your logs for any Krisp-related error messages
  • Confirm that you have not added any Krisp-related dependencies to your requirements.txt or pyproject.toml files, as they are not needed and may cause conflicts.