Echo Cancellation in Meeting Software: How It Works

Echo cancellation is one of the unsung heroes of modern video conferencing. Without it, every meeting would be plagued by feedback loops, delayed audio reflections, and garbled speech. For meeting transcription tools, echo cancellation is even more critical — poor echo handling leads to duplicate transcriptions and reduced accuracy.

What Causes Echo in Meetings?

Acoustic echo occurs when your microphone picks up audio from your speakers. In a typical remote meeting scenario:

1. The remote participant speaks
2. Their audio plays through your speakers
3. Your microphone captures that audio along with your voice
4. The mixed signal is sent back to the remote participant
5. They hear their own voice with a delay — the echo

Acoustic Echo Cancellation (AEC) Algorithms

AEC works by modeling the acoustic path between your speakers and microphone, then subtracting the predicted echo from the microphone signal:

# Simplified AEC concept
import numpy as np

class SimpleAEC:
    def __init__(self, filter_length=1024, step_size=0.01):
        self.w = np.zeros(filter_length)  # Adaptive filter
        self.mu = step_size

    def process(self, reference, microphone):
        """
        reference: audio being played (what we want to cancel)
        microphone: raw microphone input (contains echo + desired speech)
        """
        # Predict echo using adaptive filter
        echo_estimate = np.convolve(reference, self.w, mode='full')[:len(microphone)]

        # Subtract estimated echo
        clean_signal = microphone - echo_estimate

        # Update filter (Normalized LMS)
        error = clean_signal
        norm = np.dot(reference, reference) + 1e-10
        self.w += self.mu * error * reference / norm

        return clean_signal

Why AEC Matters for Transcription

Without echo cancellation, a transcription tool capturing system audio would transcribe everything twice — once when it's spoken and once when the echo is picked up. This creates duplicate, garbled transcripts that confuse both humans and AI systems.

Modern AEC Approaches

1. WebRTC-Based AEC

Most video conferencing tools use WebRTC's built-in AEC, which handles echo cancellation before audio reaches the application. This works well for communication but doesn't help with system audio capture.

2. Deep Learning AEC

Neural network-based AEC models can handle non-linear echo (caused by speaker distortion) that traditional algorithms struggle with. These models are trained on thousands of hours of echo-contaminated audio and achieve superior performance in challenging acoustic environments.

3. Reference-Signal Cancellation

For tools that capture both system audio (what others say) and microphone audio (what you say), the system audio serves as a perfect reference signal. This reference-based approach achieves near-perfect echo cancellation because the exact signal to be canceled is known.

Practical Tips for Users

• Use headphones: The single most effective way to eliminate echo. If your speakers and microphone are physically separated (headphones), there's minimal echo to cancel.
• Reduce speaker volume: Lower volume means less echo energy for the AEC to handle.
• Use a directional microphone: Directional mics pick up less speaker spillover than omnidirectional ones.
• Avoid hard surfaces: Rooms with hard walls and floors create more reflections, making AEC harder.

"Echo cancellation is the foundation of good meeting audio. Get it right, and everything downstream — transcription, AI analysis, summarization — works dramatically better." — Audio Engineer, Voxclar

For more on the meeting audio pipeline, read our WASAPI audio capture guide and real-time speech-to-text guide.