Skip to content

Request: Training dataset and model architecture details for silero_vad.onnx/.jit #780

Description

Hi Silero team,

We're using the pre-trained silero_vad.onnx model (via the android-vad wrapper) in our project and would like to properly document/attribute the model internally.

We've reviewed:

src/silero_vad/model.py — confirms this loads a pre-compiled .jit/.onnx artifact, but doesn't define the architecture.
The Quality-Metrics wiki page — lists evaluation/benchmark datasets (ESC-50, AliMeeting, Earnings21, MSDWild, AISHELL-4, VoxConverse, Libriparty, etc.), but these appear to be test/eval sets rather than the training data.
Could you help clarify:

What dataset(s) were used to train the released VAD model (as opposed to the evaluation sets listed on the Quality-Metrics page)?
Is the model architecture (layer structure/network design) documented anywhere publicly, or is it considered proprietary?

Metadata

Metadata

Assignees

Labels

help wantedExtra attention is needed

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions