A multilingual, long-form, low-SNR evaluation set introduced in the paper “DPDFNet: Boosting DeepFilterNet2 via Dual-Path RNN”.
12 languages: arabic, dutch, french, german, hungarian, korean, polish, portuguese, russian, spanish, turkish, vietnamese
9 noise scenes: airport, car, office, pub, rain, restaurant, street, subway, train
SNR (dB): 0 / 5 / 10
Duration: ~2.5 minutes per clip (includes speech-free intervals)
Size: 324 clips… See the full description on the dataset page:
https://huggingface.co/datasets/Ceva-IP/DPDFNet_EvalSet.