search

XinleiRen / MTFAA-Net

An unofficial non-causal Tensorflow implementation of "Multi-Scale Temporal Frequency Convolutional Network With Axial Attention for Speech Enhancement"
12 stars 5 forks source link

Unexpected error #1

Open shawnyxf opened 1 year ago

shawnyxf commented 1 year ago

When i run the training process of your code, i got an error "tensorflow.python.framework.errors_impl.InvalidArgumentError: num_lower must be negative or less or equal to number of rows (0) got: 150"

Can u help me figure it out? thanks in advance

XinleiRen commented 1 year ago

Hi, the error is from this function: asam.py/asat(), and the kernel function is tf.linalg.band_part(). The frame number of audios which is fed into the model is 200 (you can see it in mtfaa_net.py). And I guess you changed the frame number to a smaller value. So you can:

  1. use the default value of frame number (i.e. 200), that is you need feed a longer audio to the model;
  2. change the default value of a_frames(which is 300 now) to a smaller value to satisfy your frame number;

Thanks.