Skip to content

ReDimNet2 ONNX

Run the official PalabraAI ReDimNet2 speaker-verification models with ONNX Runtime.

All 20 published checkpoints are available and produce L2-normalized 192-dimensional speaker embeddings.

Quick start

For raw 16 kHz audio on CPU:

pip install "redimnet2-onnx[cpu,waveform]"

from redimnet2_onnx import load_model

model = load_model("b6-vb2+vox2_v0-lm")

# waveform: mono 16 kHz audio, 1-30 seconds
embedding = model.embed(waveform)

print(embedding.shape)
# (1, 192)

Models are downloaded from versioned GitHub Releases and verified with SHA-256.

Next steps

  • Installation - choose an inference backend.
  • Usage - waveform inference, speaker verification, providers, and TensorRT.
  • Models - understand model names and released checkpoint families.
  • API Reference - complete Python API.