Copy page

NVnvidia/Frame_VAD_Multilingual_MarbleNet_v2.0

NeMoVAD

Runs multilingual MarbleNet frame VAD and retains frame scores for inspection.

Voice activity detectionVoiceHub-nativemarblenet-vadParameters: Not reportedNot text-language conditionedTraining: nativeLicense: Checkpoint-specific

Parameter metadata: Not reported: the audited metadata available for the registered default does not provide an exact parameter total.

Use this model Copy model ID Checkpoint

Resources

Paper Upstream GitHub VoiceHub source Open in Colab

Usage

Complete the VoiceHub installation once, then run this repository-authored example. Model pages intentionally contain no package-install command.

This example is maintained against VoiceHub's public API; it is not copied from an upstream demo or package README.

Model-specific path: Runs multilingual MarbleNet frame VAD and retains frame scores for inspection.

Inputs and controls: Frame scores help calibration but are not speaker labels or ASR confidence values.

from pathlib import Path

from voicehub import AutoModelForVoiceActivityDetection

AUDIO_FILE = Path("speech.wav")
if not AUDIO_FILE.is_file():
    raise FileNotFoundError(AUDIO_FILE)

model = AutoModelForVoiceActivityDetection.from_pretrained(
    'nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0',
    model_type='vad_nemo',
    device="cpu",
    lazy_load=True,
)
output = model.detect(
    AUDIO_FILE,
    threshold=0.5,
    min_silence_duration_ms=200,
    return_frames=True,
)
for segment in output.segments:
    print(segment.start, segment.end, segment.score)

Use authorized recordings. Verify hardware needs and pin a revision in production.

Overview

vad_nemo is a VoiceHub voice activity detection integration. This page is generated from its registry contract. Open the vad_nemo Colab notebook.

Property	Value
Task	Voice activity detection
Architecture	marblenet-vad
Runtime	VoiceHub-native
Languages	Not text-language conditioned
Capabilities	voice-activity-detection, voicehub-native, safetensors, trusted-checkpoint-conversion, frame-scores, fine-tuning
Reusable components	—
Normalized output	VADOutput

Language support

The public VAD contract does not select a spoken language; validate checkpoint acoustic coverage on the target languages and recording conditions.

Paper and GitHub

• Paper: MarbleNet: Deep 1D Time-Channel Separable Convolutional Neural Network for VAD
• Upstream GitHub: NVIDIA NeMo
• VoiceHub source: VoiceHub model implementation

Configuration

Load configuration without constructing the model:

from voicehub import AutoConfig

config = AutoConfig.for_model('vad_nemo')
print(config.model_type)

Property	Value
Canonical model type	vad_nemo
Configuration class	NeMoVADConfig
Architecture class	NeMoVADForVoiceActivityDetection

Processing

Create the registered processor without allocating model weights:

from voicehub import AutoProcessor

processor = AutoProcessor.from_pretrained(
    'nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0',
    model_type='vad_nemo',
)
print(type(processor).__name__)

Inference

The Usage example returns VADOutput through AutoModelForVoiceActivityDetection.

Input and output contract

Property	Value
Label boundary	Clip-, frame-, or segment-level labels
Required training inputs	waveforms, labels

Use authorized audio and preserve annotation provenance. See the ASR and VAD data workflow.

Training and optimization

Use available_optimization_passes() to discover reversible public passes. Unsupported runtime or hardware fails closed before mutation.

Training contract

Property	Value
Support	native
Family	frame-classification
Recipe	single-phase
Default phase	voice_activity_detection
Training checkpoint	nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0
Native training graph	yes

Phase	Kind	Components	Required inputs	Loss keys
voice_activity_detection	objective	model	waveforms, labels	loss

The integration accepts its declared source or prepared contract directly. Call model.validate_training_support() first, then follow the training workflow.

Checkpoints, provenance, license, and limitations

Property	Value
Default checkpoint	nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0
Hugging Face ID	nvidia/Frame_VAD_Multilingual_MarbleNet_v2.0 Repository availability verified through the Hugging Face model API on 2026-08-11; pin a revision before production use.
Checkpoint status	Registry default; pin an immutable revision for production and reproducible evidence
Optional dependency extra	Core package
Hardware and runtime	Usage selects cpu; verify checkpoint-specific requirements
Real-checkpoint evidence	Release evidence; a registry default alone is not execution evidence
Implementation	voicehub.models.vad_nemo.modeling_vad_nemo.NeMoVADForVoiceActivityDetection
Configuration	voicehub.models.vad_nemo.configuration_vad_nemo.NeMoVADConfig
Source provenance	voicehub/architectures/marblenet_vad/SOURCE.json
License	Checkpoint-specific

No VoiceHub-specific license override is registered. Verify the checkpoint and upstream source terms before use.

Confirm the checkpoint revision, access terms, provenance, and license.

Limitations

• No integration-specific checkpoint limitation is registered. Verify the selected checkpoint revision and its documented runtime requirements.
• Validate memory, precision, and optional dependencies on the target system.
• Public optimizations fail closed when the runtime or hardware cannot satisfy their validation contract; an unavailable pass is not reported as applied.
• Contract tests do not replace the linked released-checkpoint evidence.

Public API

Use the stable configuration, processor, and task-model facades below.

Configuration

NeMoVADConfig

View source

NeMoVADConfig(**config_kwargs)

Parameters

• **config_kwargs — Configuration fields validated by NeMoVADConfig.

Model

NeMoVADForVoiceActivityDetection

View source

AutoModelForVoiceActivityDetection.from_pretrained(
    pretrained_model_name_or_path,
    *,
    model_type='vad_nemo',
    config=None,
    **model_kwargs,
)

Parameters

• pretrained_model_name_or_path — Hub ID or compatible local directory.
• model_type — Canonical model type; use 'vad_nemo'.
• config — Optional preloaded NeMoVADConfig instance.
• **model_kwargs — Model-specific loading arguments.

from voicehub import get_model_spec

spec = get_model_spec('vad_nemo')
print(spec.display_name, spec.task.value)

Purpose	Public object
Discover	get_model_spec('vad_nemo')
Load and run	AutoModelForVoiceActivityDetection
Configure	NeMoVADConfig
Process	AutoProcessor
Model implementation	NeMoVADForVoiceActivityDetection
Normalized output	VADOutput
Training contract	get_training_spec('vad_nemo')
Optimization lifecycle	available_optimization_passes, apply_optimization_plan, optimization_manifest, restore_optimization_plan

See all model guides, inference, and the training matrix.