remove debug info and format code

examples/librispeech/s1/conf/preprocess.yaml

@@ -10,16 +10,16 @@ process:
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument
   - type: time_warp
-    max_time_warp: 0
+    max_time_warp: 5
     inplace: true
     mode: PIL
   - type: freq_mask
-    F: 10
+    F: 30
     n_mask: 2
     inplace: true
-    replace_with_zero: true
+    replace_with_zero: false
   - type: time_mask
-    T: 50
+    T: 40
     n_mask: 2
     inplace: true
-    replace_with_zero: true
+    replace_with_zero: false

paddlespeech/s2t/frontend/audio.py

@@ -24,9 +24,9 @@ import soundfile
 import soxbindings as sox
 from scipy import signal
 
-from .utility import subfile_from_tar
-from .utility import convert_samples_to_float32
 from .utility import convert_samples_from_float32
+from .utility import convert_samples_to_float32
+from .utility import subfile_from_tar
 
 
 class AudioSegment():

paddlespeech/s2t/frontend/utility.py

@@ -390,4 +390,3 @@ def convert_samples_from_float32(samples, dtype):
     else:
         raise TypeError("Unsupported sample type: %s." % samples.dtype)
     return output_samples.astype(dtype)
-

paddlespeech/s2t/transform/spec_augment.py

@@ -34,6 +34,9 @@ def time_warp(x, max_time_warp=80, inplace=False, mode="PIL"):
     :returns numpy.ndarray: time warped spectrogram (time, freq)
     """
     window = max_time_warp
+    if window == 0:
+        return x
+
     if mode == "PIL":
         t = x.shape[0]
         if t - window <= window:

paddlespeech/s2t/transform/spectrogram.py

@@ -307,9 +307,6 @@ class IStft():
             center=self.center, )
 
 
-from paddlespeech.s2t.utils.log import Log
-logger = Log(__name__).getlog()
-
 class LogMelSpectrogramKaldi():
     def __init__(
             self,
@@ -347,22 +344,22 @@ class LogMelSpectrogramKaldi():
         self.dither = dither
 
     def __repr__(self):
-        return ("{name}(fs={fs}, n_mels={n_mels}, n_fft={n_fft}, "
-                "n_shift={n_shift}, win_length={win_length}, window={window}, "
-                "fmin={fmin}, fmax={fmax}, eps={eps}, preemph={preemph}, window={window}, dither={dither}))".format(
+        return (
+            "{name}(fs={fs}, n_mels={n_mels}, n_fft={n_fft}, "
+            "n_shift={n_shift}, win_length={win_length}, preemph={preemph}, window={window}, "
+            "fmin={fmin}, fmax={fmax}, eps={eps}, dither={dither}))".format(
                 name=self.__class__.__name__,
                 fs=self.fs,
                 n_mels=self.n_mels,
                 n_fft=self.n_fft,
                 n_shift=self.n_shift,
+                preemph=self.preemph,
                 win_length=self.win_length,
                 window=self.window,
                 fmin=self.fmin,
                 fmax=self.fmax,
                 eps=self.eps,
-                    preemph=self.preemph,
-                    window=self.window,
-                    dither=self.dither))
+                dither=self.dither, ))
 
     def __call__(self, x):
         """
@@ -379,12 +376,10 @@ class LogMelSpectrogramKaldi():
         if x.ndim != 1:
             raise ValueError("Not support x: [Time, Channel]")
 
-        logger.info(f"in {x}")
         if x.dtype in np.sctypes['float']:
             # PCM32 -> PCM16
             bits = np.iinfo(np.int16).bits
             x = x * 2**(bits - 1)
-        logger.info(f"b {x}")
 
         # logfbank need PCM16 input
         y = logfbank(
@@ -400,7 +395,4 @@ class LogMelSpectrogramKaldi():
             remove_dc_offset=self.remove_dc_offset,
             preemph=self.preemph,
             wintype=self.window)
-        logger.info(f"a {y}")
-
-
         return y