Tensorflow api for Delta and acceleration features Computation

General Discussion
,
#1

Hi,

Is there any tensorflow api or any method to compute delta and acceleration features for log mel filterbank energies?

I tried this way…
Converted numpy delta computing code to tensorflow code by using tf apis. I’m passing log mel features with delta and acceleration features to my model. But it is throwing error " RecursionError: maximum recursion depth exceeded" may be because of tf.concat layer.

def delta_compute(feat, N):
       if N < 1:
         raise ValueError('N must be an integer >= 1')

       NUMFRAMES = feat.shape[0]
       denominator = 2 * sum([i**2 for i in range(1, N+1)])
       frame_feat = tf.zeros((1,feat.shape[1]))
       padded = tf.pad( feat[p], [ [ N, N ], [ 0, 0 ] ], "SYMMETRIC" )
       for t in range(NUMFRAMES):
               res = tf.tensordot(tf.range(-N,N+1,delta =1,dtype=tf.float32), padded[t : t+2*N+1],1) / denominator
               frame_feat = tf.concat([frame_feat,tf.reshape(res,(1,feat.shape[1]))],0)
       return frame_feat[1:]

Is there any tensorflow api to directly compute delta features by giving log fbank energies? Or otherwise how to fix the error caused by tf.concat layer.

Thanks for the help!!