Version bump

docs/source/changelog.rst

@@ -6,6 +6,20 @@
 Changelog
 =========
 
+2.0.0a12
+--------
+
+- Updated how sample rates are handled. MFA now generates features between 80 Hz and 7800 Hz and allows downsampling and
+  upsampling, so there will be no more errors or warnings about unsupported sample rates or speakers with different sample
+  rates
+- Fixed a bug where some options for generating MFCCs weren't properly getting picked up (e.g., snip-edges)
+- (EXPERIMENTAL) Added better support for varying frame shift. In :code:`mfa align`, you can now add a flag of :code:`--frame_shift 1` to align
+  with millisecond shifts between frames.  Please note this is more on the experimental side, as it increases computational
+  time significantly and I don't know fully the correct options to use for :code:`self_loop_scale`, :code:`transition_scale`,
+  and :code:`acoustic_scale` to generate good alignments.
+- Fixed a bug in G2P training with relative paths for output model
+- Cleaned up validator output
+
 2.0.0a11
 --------
 

montreal_forced_aligner/__init__.py

@@ -1,6 +1,6 @@
 __ver_major__ = 2
 __ver_minor__ = 0
-__ver_patch__ = '0a11'
+__ver_patch__ = '0a12'
 __version__ = "{}.{}.{}".format(__ver_major__, __ver_minor__, __ver_patch__)
 
 __all__ = ['aligner', 'command_line', 'models', 'corpus', 'config', 'dictionary', 'exceptions',

montreal_forced_aligner/corpus/align_corpus.py

@@ -219,6 +219,7 @@ def _load_from_source_mp(self):
                 for w in words:
                     new_w = re.split(r"[-']", w)
                     self.word_counts.update(new_w + [w])
+                self.wav_files.append(file_name)
                 self.text_mapping[utt_name] = ' '.join(words)
                 self.utt_text_file_mapping[utt_name] = info['text_file']
                 self.speak_utt_mapping[speaker_name].append(utt_name)

montreal_forced_aligner/validator.py

@@ -149,10 +149,6 @@ class CorpusValidator(object):
     UNREADABLE TEXT FILES
     --------------------
     {}
-    
-    UNSUPPORTED SAMPLE RATES
-    --------------------
-    {}
     '''
 
     alignment_analysis_template = '''
@@ -210,8 +206,7 @@ def analyze_setup(self):
                                                               self.analyze_files_with_no_transcription(),
                                                               self.analyze_transcriptions_with_no_wavs(),
                                                               self.analyze_textgrid_read_errors(),
-                                                              self.analyze_unreadable_text_files(),
-                                                              self.analyze_unsupported_sample_rates()
+                                                              self.analyze_unreadable_text_files()
                                                               ))
 
     def analyze_oovs(self):
@@ -340,23 +335,6 @@ def analyze_unreadable_text_files(self):
             message = 'There were no issues reading text files.'
         return message
 
-    def analyze_unsupported_sample_rates(self):
-        output_dir = self.corpus.output_directory
-        if self.corpus.unsupported_sample_rate:
-            path = os.path.join(output_dir, 'unsupported_sample_rates.csv')
-            with open(path, 'w') as f:
-                for file_path in self.corpus.unsupported_sample_rate:
-                    f.write('{}\n'.format(file_path))
-            message = 'There were {} sound files with sample rates <16000. ' \
-                      'Feature generation targets from 20 Hz to 7800 Hz, ' \
-                      'so lower sample rates may produce malformed features. ' \
-                      'These feature might still work, particularly when not using ' \
-                      'an existing acoustic model, but be aware of potential issues.' \
-                      'Please see {} for a list.'.format(len(self.corpus.unsupported_sample_rate), path)
-        else:
-            message = 'There were no sound files with unsupported sample rates.'
-        return message
-
     def analyze_unaligned_utterances(self):
         unaligned_utts = self.trainer.get_unaligned_utterances()
         num_utterances = self.corpus.num_utterances