About using BEATs as audio feature extractor about unilm

[tcourat]

I belive that the model already preprocess input data to 16000 Hz here :

unilm/beats/BEATs.py

Line 127 in 8ee6f74

fbank = ta_kaldi.fbank(waveform, num_mel_bins=128, sample_frequency=16000, frame_length=25, frame_shift=10)

[XiaokangY]

I belive that the model already preprocess input data to 16000 Hz here :我相信该模型已经将输入数据预处理为 16000 Hz：

unilm/beats/BEATs.py

Line 127 in 8ee6f74

fbank = ta_kaldi.fbank(waveform, num_mel_bins=128, sample_frequency=16000, frame_length=25, frame_shift=10)

Haha, I have solved this problem. The problem is that the loss cannot be reduced when the following features are input into the pre-trained model for training. This problem has been bothering me for a long time.