silero

Silero Voice Activity Detection (VAD) implementation for Pipecat.

This module provides a VAD analyzer based on the Silero VAD ONNX model, which can detect voice activity in audio streams with high accuracy. Supports 8kHz and 16kHz sample rates.

class pipecat.audio.vad.silero.SileroOnnxModel(path, force_onnx_cpu=True)[source]

Bases: object

ONNX runtime wrapper for the Silero VAD model.

Provides voice activity detection using the pre-trained Silero VAD model with ONNX runtime for efficient inference. Handles model state management and input validation for audio processing.

__init__(path, force_onnx_cpu=True)[source]

Initialize the Silero ONNX model.

Parameters:
  • path – Path to the ONNX model file.

  • force_onnx_cpu – Whether to force CPU execution provider.

reset_states(batch_size=1)[source]

Reset the internal model states.

Parameters:

batch_size – Batch size for state initialization. Defaults to 1.

class pipecat.audio.vad.silero.SileroVADAnalyzer(*, sample_rate: int | None = None, params: VADParams | None = None)[source]

Bases: VADAnalyzer

Voice Activity Detection analyzer using the Silero VAD model.

Implements VAD analysis using the pre-trained Silero ONNX model for accurate voice activity detection. Supports 8kHz and 16kHz sample rates.

The model’s recurrent state is reset about every two seconds of audio, but only during clear non-speech. Non-speech such as music, or long silence at 8kHz, leaves the state unable to detect the speech that follows until it is reset. A reset during speech makes the model briefly lose confidence, which can end the user’s speech early, so a due reset waits until the user is quiet.

__init__(*, sample_rate: int | None = None, params: VADParams | None = None)[source]

Initialize the Silero VAD analyzer.

Parameters:
  • sample_rate – Audio sample rate (8000 or 16000 Hz). If None, will be set later.

  • params – VAD parameters for detection thresholds and timing.

set_sample_rate(sample_rate: int)[source]

Set the sample rate for audio processing.

Parameters:

sample_rate – Audio sample rate (must be 8000 or 16000 Hz).

Raises:

ValueError – If sample rate is not 8000 or 16000 Hz.

num_frames_required() → int[source]

Get the number of audio frames required for VAD analysis.

Returns:

Number of frames required (512 for 16kHz, 256 for 8kHz).

voice_confidence(buffer) → float[source]

Calculate voice activity confidence for the given audio buffer.

Parameters:

buffer – Audio buffer to analyze.

Returns:

Voice confidence score between 0.0 and 1.0.