[asr] logfbank with dither (#1179)

* fix logfbank dither

* format

examples/aishell/asr1/conf/conformer.yaml

@@ -3,14 +3,7 @@ data:
   train_manifest: data/manifest.train
   dev_manifest: data/manifest.dev
   test_manifest: data/manifest.test
-  min_input_len: 0.5
-  max_input_len: 20.0 # second
-  min_output_len: 0.0
-  max_output_len: 400.0
-  min_output_input_ratio: 0.05
-  max_output_input_ratio: 10.0
-
-
+  
 collator:
   vocab_filepath: data/lang_char/vocab.txt 
   unit_type: 'char'

examples/aishell/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

examples/aishell/asr1/conf/transformer.yaml

@@ -73,7 +73,7 @@ model:
 
 
 training:
-  n_epoch: 120 
+  n_epoch: 240 
   accum_grad: 2
   global_grad_clip: 5.0
   optim: adam

examples/callcenter/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

examples/librispeech/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

examples/other/g2p/get_g2p_data.py

@@ -28,7 +28,8 @@ def get_baker_data(root_dir):
     alignment_files = [f for f in alignment_files if f.stem not in exclude]
     data_dict = defaultdict(dict)
     for alignment_fp in alignment_files:
-        alignment = textgrid.openTextgrid(alignment_fp, includeEmptyIntervals=True)
+        alignment = textgrid.openTextgrid(
+            alignment_fp, includeEmptyIntervals=True)
         # only with baker's annotation
         utt_id = alignment.tierNameList[0].split(".")[0]
         intervals = alignment.tierDict[alignment.tierNameList[0]].entryList

examples/timit/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

examples/tiny/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

examples/wenetspeech/asr1/conf/preprocess.yaml

@@ -5,7 +5,7 @@ process:
     n_mels: 80
     n_shift: 160
     win_length: 400
-    dither: true
+    dither: 0.1
   - type: cmvn_json
     cmvn_path: data/mean_std.json
   # these three processes are a.k.a. SpecAugument

paddlespeech/s2t/models/u2/u2.py

@@ -584,8 +584,9 @@ class U2BaseModel(ASRInterface, nn.Layer):
             hyp_content = hyp[0]
             # Prevent the hyp is empty
             if len(hyp_content) == 0:
-                hyp_content = (self.ctc.blank_id,)
-            hyp_content = paddle.to_tensor(hyp_content, place=device, dtype=paddle.long)
+                hyp_content = (self.ctc.blank_id, )
+            hyp_content = paddle.to_tensor(
+                hyp_content, place=device, dtype=paddle.long)
             hyp_list.append(hyp_content)
         hyps_pad = pad_sequence(hyp_list, True, self.ignore_id)
         hyps_lens = paddle.to_tensor(

paddlespeech/s2t/transform/spectrogram.py

@@ -319,7 +319,7 @@ class LogMelSpectrogramKaldi():
             fmin=20,
             fmax=None,
             eps=1e-10,
-            dither=False):
+            dither=1.0):
         self.fs = fs
         self.n_mels = n_mels
         self.n_fft = n_fft
@@ -374,7 +374,7 @@ class LogMelSpectrogramKaldi():
         Returns:
             np.ndarray: (T, D)
         """
-        dither = self.dither if train else False
+        dither = self.dither if train else 0.0
         if x.ndim != 1:
             raise ValueError("Not support x: [Time, Channel]")
 

paddlespeech/t2s/exps/gan_vocoder/hifigan/train.py

@@ -242,8 +242,7 @@ def train_sp(args, config):
 def main():
     # parse args and config and redirect to train_sp
 
-    parser = argparse.ArgumentParser(
-        description="Train a HiFiGAN model.")
+    parser = argparse.ArgumentParser(description="Train a HiFiGAN model.")
     parser.add_argument(
         "--config", type=str, help="config file to overwrite default config.")
     parser.add_argument("--train-metadata", type=str, help="training data.")

tools/release_note.py

@@ -14,191 +14,180 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # ==============================================================================
-
 """ Create release notes with the issues from a milestone.
     python3 release_notes.py -c didi delta v.xxxxx
 """
-
-import sys
-import json
 import argparse
-import urllib.request
 import collections
+import json
+import sys
+import urllib.request
 
 github_url = 'https://api.github.com/repos'
 
 if __name__ == '__main__':
-  # usage: 
-  # 1. close milestone on github
-  # 2. python3 tools/release_notes.py -c didi delta v0.3.3
-
-  # Parse arguments
-  parser = argparse.ArgumentParser(
-    description='Create a draft release with the issues from a milestone.',
-  )
-
-  parser.add_argument(
-    'user',
-    metavar='user',
-    type=str,
-    default='paddlepaddle',
-    help='github user: paddlepaddle'
-  )
-
-  parser.add_argument(
-    'repository',
-    metavar='repository',
-    type=str,
-    default='paddlespeech',
-    help='github repository: paddlespeech'
-  )
-
-  parser.add_argument(
-    'milestone',
-    metavar='milestone',
-    type=str,
-    help='name of used milestone: v0.3.3'
-  )
-
-  parser.add_argument(
-    '-c', '--closed',
-    help='Fetch closed milestones/issues',
-    action='store_true'
-  )
-
-  parser.print_help()
-  args = parser.parse_args()
-
-  # Fetch milestone infos
-  url = "%s/%s/%s/milestones" % (
-    github_url,
-    args.user,
-    args.repository
-  )
-
-  headers = {
-    'Origin': 'https://github.com',
-    'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64) '
-                  'AppleWebKit/537.11 (KHTML, like Gecko) '
-                  'Chrome/23.0.1271.64 Safari/537.11',
-    'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
-    'Accept-Charset': 'ISO-8859-1,utf-8;q=0.7,*;q=0.3',
-    'Accept-Encoding': 'none',
-    'Accept-Language': 'en-US,en;q=0.8',
-    'Connection': 'keep-alive'}
-
-  if args.closed:
-    url += "?state=closed"
-
-  req = urllib.request.Request(url, headers=headers)
-  github_request = urllib.request.urlopen(req)
-  if not github_request:
-    parser.error('Cannot read milestone list.')
-
-  decoder = json.JSONDecoder()
-  milestones = decoder.decode(github_request.read().decode('utf-8'))
-  github_request.close()
-
-  print('parse milestones', file=sys.stderr)
-  milestone_id = None
-  for milestone in milestones:
-    if milestone['title'] == args.milestone:
-      milestone_id = milestone['number']
-  if not milestone_id:
-    parser.error('Cannot find milestone')
-
-
-  # Get milestone related issue info
-  url = '%s/%s/%s/issues?milestone=%d' % (
-    github_url,
-    args.user,
-    args.repository,
-    milestone_id
-  )
-  if args.closed:
-    url += "&state=closed"
-
-  req = urllib.request.Request(url, headers=headers)
-  github_request = urllib.request.urlopen(req)
-  if not github_request:
-    parser.error('Cannot read issue list.')
-
-  issues = decoder.decode(github_request.read().decode('utf-8'))
-  github_request.close()
-
-  #print('parse issues', file=sys.stderr)
-  #final_data = []
-  #labels = []
-  #thanks_to = []
-  #for issue in issues:
-
-  #  for label in issue['labels']:
-  #    labels.append(label['name'])
-
-  #  thanks_to.append('@%s' % (issue['user']['login']))
-  #  final_data.append(' * **[%s]** - %s #%d by **@%s**\n' % (
-  #    label['name'],
-  #    issue['title'],
-  #    issue['number'],
-  #    issue['user']['login']
-  #  ))
-
-  #dic = collections.defaultdict(set)
-  #for l_release in list(set(labels)):
-
-  #  for f_data in final_data:
-  #    if '[%s]' % l_release in f_data:
-  #      dic[l_release].add(f_data)
-
-  #with open(f"release_note_issues_{args.milestone}.md", 'w') as f:
-  #  for key, value in dic.items():
-  #    print('# %s\n%s' % (key, ''.join(value)), file=f)
-  #  print('# %s\n%s' % ('Acknowledgements', 'Special thanks to %s ' % ('  '.join(list(set(thanks_to))))), file=f)
-
-
-  # Get milestone related PR info
-  url = '%s/%s/%s/pulls?milestone=%d' % (
-    github_url,
-    args.user,
-    args.repository,
-    milestone_id
-  )
-  if args.closed:
-    url += "&state=closed"
-
-  req = urllib.request.Request(url, headers=headers)
-  github_request = urllib.request.urlopen(req)
-  if not github_request:
-    parser.error('Cannot read issue list.')
-
-  issues = decoder.decode(github_request.read().decode('utf-8'))
-  github_request.close()
-
-  print('parse pulls', file=sys.stderr)
-  final_data = []
-  labels = []
-  thanks_to = []
-  for issue in issues:
-
-    for label in issue['labels']:
-      labels.append(label['name'])
-
-    thanks_to.append('@%s' % (issue['user']['login']))
-    final_data.append(' * **[%s]** - %s #%d by **@%s**\n' % (
-      label['name'],
-      issue['title'],
-      issue['number'],
-      issue['user']['login']
-    ))
-
-  dic = collections.defaultdict(set)
-  for l_release in list(set(labels)):
-
-    for f_data in final_data:
-      if '[%s]' % l_release in f_data:
-        dic[l_release].add(f_data)
-
-  with open(f"release_note_pulls_{args.milestone}.md", 'w') as f:
-    for key, value in dic.items():
-      print('# %s\n%s' % (key, ''.join(value)), file=f)
-    print('# %s\n%s' % ('Acknowledgements', 'Special thanks to %s ' % ('  '.join(list(set(thanks_to))))), file=f)
+    # usage: 
+    # 1. close milestone on github
+    # 2. python3 tools/release_notes.py -c didi delta v0.3.3
+
+    # Parse arguments
+    parser = argparse.ArgumentParser(
+        description='Create a draft release with the issues from a milestone.',
+    )
+
+    parser.add_argument(
+        'user',
+        metavar='user',
+        type=str,
+        default='paddlepaddle',
+        help='github user: paddlepaddle')
+
+    parser.add_argument(
+        'repository',
+        metavar='repository',
+        type=str,
+        default='paddlespeech',
+        help='github repository: paddlespeech')
+
+    parser.add_argument(
+        'milestone',
+        metavar='milestone',
+        type=str,
+        help='name of used milestone: v0.3.3')
+
+    parser.add_argument(
+        '-c',
+        '--closed',
+        help='Fetch closed milestones/issues',
+        action='store_true')
+
+    parser.print_help()
+    args = parser.parse_args()
+
+    # Fetch milestone infos
+    url = "%s/%s/%s/milestones" % (github_url, args.user, args.repository)
+
+    headers = {
+        'Origin':
+        'https://github.com',
+        'User-Agent':
+        'Mozilla/5.0 (X11; Linux x86_64) '
+        'AppleWebKit/537.11 (KHTML, like Gecko) '
+        'Chrome/23.0.1271.64 Safari/537.11',
+        'Accept':
+        'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
+        'Accept-Charset':
+        'ISO-8859-1,utf-8;q=0.7,*;q=0.3',
+        'Accept-Encoding':
+        'none',
+        'Accept-Language':
+        'en-US,en;q=0.8',
+        'Connection':
+        'keep-alive'
+    }
+
+    if args.closed:
+        url += "?state=closed"
+
+    req = urllib.request.Request(url, headers=headers)
+    github_request = urllib.request.urlopen(req)
+    if not github_request:
+        parser.error('Cannot read milestone list.')
+
+    decoder = json.JSONDecoder()
+    milestones = decoder.decode(github_request.read().decode('utf-8'))
+    github_request.close()
+
+    print('parse milestones', file=sys.stderr)
+    milestone_id = None
+    for milestone in milestones:
+        if milestone['title'] == args.milestone:
+            milestone_id = milestone['number']
+    if not milestone_id:
+        parser.error('Cannot find milestone')
+
+    # Get milestone related issue info
+    url = '%s/%s/%s/issues?milestone=%d' % (github_url, args.user,
+                                            args.repository, milestone_id)
+    if args.closed:
+        url += "&state=closed"
+
+    req = urllib.request.Request(url, headers=headers)
+    github_request = urllib.request.urlopen(req)
+    if not github_request:
+        parser.error('Cannot read issue list.')
+
+    issues = decoder.decode(github_request.read().decode('utf-8'))
+    github_request.close()
+
+    #print('parse issues', file=sys.stderr)
+    #final_data = []
+    #labels = []
+    #thanks_to = []
+    #for issue in issues:
+
+    #  for label in issue['labels']:
+    #    labels.append(label['name'])
+
+    #  thanks_to.append('@%s' % (issue['user']['login']))
+    #  final_data.append(' * **[%s]** - %s #%d by **@%s**\n' % (
+    #    label['name'],
+    #    issue['title'],
+    #    issue['number'],
+    #    issue['user']['login']
+    #  ))
+
+    #dic = collections.defaultdict(set)
+    #for l_release in list(set(labels)):
+
+    #  for f_data in final_data:
+    #    if '[%s]' % l_release in f_data:
+    #      dic[l_release].add(f_data)
+
+    #with open(f"release_note_issues_{args.milestone}.md", 'w') as f:
+    #  for key, value in dic.items():
+    #    print('# %s\n%s' % (key, ''.join(value)), file=f)
+    #  print('# %s\n%s' % ('Acknowledgements', 'Special thanks to %s ' % ('  '.join(list(set(thanks_to))))), file=f)
+
+    # Get milestone related PR info
+    url = '%s/%s/%s/pulls?milestone=%d' % (github_url, args.user,
+                                           args.repository, milestone_id)
+    if args.closed:
+        url += "&state=closed"
+
+    req = urllib.request.Request(url, headers=headers)
+    github_request = urllib.request.urlopen(req)
+    if not github_request:
+        parser.error('Cannot read issue list.')
+
+    issues = decoder.decode(github_request.read().decode('utf-8'))
+    github_request.close()
+
+    print('parse pulls', file=sys.stderr)
+    final_data = []
+    labels = []
+    thanks_to = []
+    for issue in issues:
+
+        for label in issue['labels']:
+            labels.append(label['name'])
+
+        thanks_to.append('@%s' % (issue['user']['login']))
+        final_data.append(' * **[%s]** - %s #%d by **@%s**\n' %
+                          (label['name'], issue['title'], issue['number'],
+                           issue['user']['login']))
+
+    dic = collections.defaultdict(set)
+    for l_release in list(set(labels)):
+
+        for f_data in final_data:
+            if '[%s]' % l_release in f_data:
+                dic[l_release].add(f_data)
+
+    with open(f"release_note_pulls_{args.milestone}.md", 'w') as f:
+        for key, value in dic.items():
+            print('# %s\n%s' % (key, ''.join(value)), file=f)
+        print(
+            '# %s\n%s' % ('Acknowledgements', 'Special thanks to %s ' %
+                          ('  '.join(list(set(thanks_to))))),
+            file=f)