I want to add noise cancellation for background noise in our voice ai agent which is in gemini v2v how we can do this

Thanks for reaching out, Abhisek. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

i want for both

Great — since you want noise cancellation for both WebRTC and SIP participants, here’s a concise, practical plan plus ready-to-use code snippets (Python and Node) to apply the correct filter per participant and avoid double-processing. Summary of what to do • Install the noise cancellation plugin: it includes BVC (for WebRTC) and BVCTelephony (for SIP). • When a participant joins, detect whether they’re WebRTC or SIP and apply: ◦

BVC()

for WebRTC participants (full-bandwidth). ◦

BVCTelephony()

for SIP/telephony participants (narrowband). • Disable any other noise-suppression on the SIP trunk (e.g., krisp) to avoid double suppression artifacts. • Test and tune on representative calls to verify STT/Gemini quality. Installation • Add the plugin package: ◦ uv (uVenv) command:

uv add "livekit-plugins-noise-cancellation~=0.2"

Python example • This shows dynamically selecting the right filter per participant when starting the agent session:

from livekit import rtc
from livekit.agents import AgentSession, room_io
from livekit.plugins import noise_cancellation

# inside your session setup / participant join handler
participant = await ctx.wait_for_participant()  # wait for participant to join

# choose filter by participant kind
filter = (
    noise_cancellation.BVCTelephony()
    if participant.kind == rtc.ParticipantKind.PARTICIPANT_KIND_SIP
    else noise_cancellation.BVC()
)

await session.start(
    room=ctx.room,
    agent=Assistant(),
    room_options=room_io.RoomOptions(
        audio_input=room_io.AudioInputOptions(noise_cancellation=filter)
    ),
)

Node (JavaScript / TypeScript) example • Equivalent logic for Node agents:

const { rtc } = require("livekit");
const { AgentSession, room_io } = require("livekit/agents");
const { noise_cancellation } = require("livekit-plugins-noise-cancellation");

// inside your participant join handler
const participant = await ctx.waitForParticipant();

const filter =
  participant.kind === rtc.ParticipantKind.PARTICIPANT_KIND_SIP
    ? new noise_cancellation.BVCTelephony()
    : new noise_cancellation.BVC();

await session.start({
  room: ctx.room,
  agent: new Assistant(),
  roomOptions: new room_io.RoomOptions({
    audioInput: new room_io.AudioInputOptions({ noiseCancellation: filter }),
  }),
});

Key tuning & testing tips • Test with noisy backgrounds representative of your users: music, café noise, overlapping voices. • Verify STT input to Gemini: confirm lower error rates, fewer false triggers, and clearer transcription. • If you hear artifacts (pumping/warbling), ensure SIP trunk-side noise suppression (krisp) is disabled — double suppression causes artifacts. • If you still see issues on specific devices or networks, capture short audio samples from those cases and compare processed vs unprocessed audio to identify problems. What I can provide next • A drop-in example for your exact agent repo or framework if you paste the relevant start/session code. • A short checklist for validating STT/Gemini quality after enabling the filters. • Troubleshooting steps and logs to collect if you see artifacts after enabling filters. Would you like the drop-in snippet adapted to a specific file in your codebase or the validation checklist next?

give me for pyhton

Thanks — below is a focused, ready-to-drop-in Python example that applies the correct noise cancellation filter for both WebRTC and SIP participants, plus install steps and quick testing tips. Installation 1. Add the noise cancellation plugin:

uv add "livekit-plugins-noise-cancellation~=0.2"

Drop-in Python example • Place this in your agent session/participant handler where you start the agent session. It chooses

BVC()

for WebRTC participants and

BVCTelephony()

for SIP participants, then starts the session with the selected filter.

from livekit import rtc
from livekit.agents import AgentSession, room_io
from livekit.plugins import noise_cancellation

class Assistant:
    # Your assistant implementation here
    pass

async def handle_participant(ctx, session: AgentSession):
    """
    ctx: context provided by your agent runtime that includes the room and participant events
    session: your AgentSession instance
    """
    # Wait for a participant to join (example API - adapt to your runtime)
    participant = await ctx.wait_for_participant()

    # Select the appropriate noise cancellation filter
    nc_filter = (
        noise_cancellation.BVCTelephony()
        if participant.kind == rtc.ParticipantKind.PARTICIPANT_KIND_SIP
        else noise_cancellation.BVC()
    )

    # Start the agent session with the chosen noise cancellation filter applied to audio input
    await session.start(
        room=ctx.room,
        agent=Assistant(),
        room_options=room_io.RoomOptions(
            audio_input=room_io.AudioInputOptions(noise_cancellation=nc_filter)
        ),
    )

Notes and best practices • Use

BVC()

for WebRTC (full-band audio) and

BVCTelephony()

for SIP (narrowband telephony). • Disable upstream SIP/trunk-side noise suppression (e.g., krisp) to avoid double-processing and artifacts. • Noise cancellation runs with very low latency (<10ms) and filters audio before STT/Gemini ingestion to improve transcription quality. • If you hear artifacts (warbling or pumping), check for double suppression and test with different background noises. Quick validation checklist • Record short samples of noisy inputs (café noise, music, overlapping voices) with and without the filter enabled. • Compare STT/transcription accuracy and error rates sent to Gemini. • Confirm no audible artifacts; if artifacts appear, ensure SIP-side suppression is disabled and re-test. If you want, I can adapt this snippet to the exact file structure of your project — paste the relevant start/session code or the function where you handle participants and I’ll give a drop-in edit.