Merge pull request #2 from HuangZiliAndy/master

Fix problem in local amplitude

Author: danpovey

lilfilter/local_amplitude.py

@@ -79,15 +79,7 @@ def __init__(self,
 
 
         self.block_size = block_size
-        if block_size > 1:
-            # num_zeros = 4 is a lower-than-normal width for the FIR filter since there
-            # won't be frequencies near the Nyquist and we don't need a sharp cutoff.
-            # filter_cutoff_ratio = 9 is to avoid aliasing effects with this less-precise
-            # filter (default is 0.95).
-            self.resampler = resampler.Resampler(block_size, num_zeros = 4,
-                                                 filter_cutoff_ratio = 0.9,
-                                                 double_precision = double_precision)
-
+        assert block_size > 1
 
     def compute(self,
                 input):
@@ -142,7 +134,13 @@ def compute(self,
         smoothed_amplitudes = self.gaussian_filter.apply(summed_amplitudes)
         assert smoothed_amplitudes.shape == summed_amplitudes.shape
 
-        upsampled_amplitudes = self.resampler.upsample(smoothed_amplitudes)
+        # num_zeros = 4 is a lower-than-normal width for the FIR filter since there
+        # won't be frequencies near the Nyquist and we don't need a sharp cutoff.
+        # filter_cutoff_ratio = 9 is to avoid aliasing effects with this less-precise
+        # filter (default is 0.95).
+        self.resampler = resampler.Resampler(1, self.block_size, dtype = self.dtype, num_zeros = 4,
+                                             cutoff_ratio = 0.9)
+        upsampled_amplitudes = self.resampler.resample(smoothed_amplitudes)
         assert upsampled_amplitudes.shape[1] >= signal_length
 
 

tests/test_local_amplitude.py

@@ -29,9 +29,9 @@ def test_constructor_and_compute(self):
         print("b.shape = {}, c.shape = {}, d.shape = {}".format(b.shape, c.shape, d.shape))
         print("b sum = ", b.sum())
         print("partial sum = ", b[0,0,5,:].sum().item())
-        plt.plot(torch.arange(b.shape[-1]), b[0,0,5,:])
+        plt.plot(torch.arange(b.shape[-1]).numpy(), b[0,0,5,:].numpy())
         print("d sum = ", d.sum())
-        plt.plot(torch.arange(d.shape[-1]), d[0,5,:])
+        plt.plot(torch.arange(d.shape[-1]).numpy(), d[0,5,:].numpy())
         plt.show()