Mixed Voice Separation Dataset — WaveFake + ESC-50 (Training & Inference Results)
收藏官方服务：
资源简介：
Audio separation dataset and results using SpecTransformerSeparator. Training data: WaveFake fake speech mixed with ESC-50 noise at SNR=5dB. Files: training_mixed.tar.gz (mixed audio inputs), training_speech.tar.gz (clean speech ground truth), training_noise.tar.gz (clean noise ground truth), infer_output.tar.gz (model separation results: est_speech, est_noise vs gt). Model: Spectrum-domain Transformer Separator (4 layers, d_model=256, 8 heads). Metrics: SI-SDR / SI-SNR.
提供机构：
Zenodo创建时间：
2026-06-06
