VOX1O-MUSAN

Materialized VoxCeleb1-O mixtures using the Music and Noise categories from MUSAN.

Structure

Music/
  -10dB/
  -5dB/
  0dB/
  5dB/
  10dB/
Noise/
  -10dB/
  -5dB/
  0dB/
  5dB/
  10dB/
manifest.json

The dataset contains 47,080 mono 16 kHz, 32-bit floating-point WAV files: 4,708 VoxCeleb1-O references for each of two MUSAN categories and five SNR levels.

Mixtures use deterministic whole-utterance power SNR with seed 1234, without clipping or peak normalization. Exact source counts, hashes, and protocol metadata are recorded in manifest.json.