Autovot support PR

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "autovot"]
+	path = autovot
+	url = https://github.com/mlml/autovot

conch/analysis/autovot.py

@@ -0,0 +1,87 @@
+from .functions import BaseAnalysisFunction
+import wave
+import subprocess
+import textgrid 
+import os
+import tempfile
+
+
+def is_autovot_friendly_file(sound_file):
+    rate = subprocess.run(["soxi", "-r", sound_file], encoding="UTF-8", stdout=subprocess.PIPE).stdout
+    if int(rate) != 16000:
+        return False
+
+    channels = subprocess.run(["soxi", "-c", sound_file], encoding="UTF-8", stdout=subprocess.PIPE).stdout
+    if int(channels) != 1:
+        return False
+    return True
+
+def resample_for_autovot(soundfile, tmpdir):
+    output_file = os.path.join(tmpdir, "sound_file.wav")
+    subprocess.call(["sox", soundfile, "-c", "1", "-r", "16000", output_file])
+    return output_file
+
+
+class MeasureVOTPretrained(object):
+    def __init__(self, classifier_to_use=None, min_vot_length=15, max_vot_length=250, window_max=30, window_min=30, debug=False):
+        if classifier_to_use is None:
+            raise ValueError("There must be a classifier to run AutoVOT")
+        else:
+            self.classifier_to_use = classifier_to_use
+        self.min_vot_length = min_vot_length
+        self.max_vot_length = max_vot_length
+        self.debug = debug
+        self.window_max = window_max
+        self.window_min = window_min
+
+    def __call__(self, segment):
+        file_path = os.path.expanduser(segment["file_path"])
+        begin = segment["begin"]
+        end = segment["end"]
+        vot_marks = sorted(segment["vot_marks"], key=lambda x: x[0])
+        grid = textgrid.TextGrid(maxTime=end)
+        vot_tier = textgrid.IntervalTier(name='vot', maxTime=end)
+        for vot_begin, vot_end, *extra_data in vot_marks:
+            vot_tier.add(vot_begin, vot_end, 'vot')
+        grid.append(vot_tier)
+        with tempfile.TemporaryDirectory() as tmpdirname:
+            grid_path = "{}/file.TextGrid".format(tmpdirname)
+            csv_path = "{}/file.csv".format(tmpdirname)
+            wav_filenames = "{}/wavs.txt".format(tmpdirname)
+            textgrid_filenames = "{}/textgrids.txt".format(tmpdirname)
+
+            if not is_autovot_friendly_file(file_path):
+                file_path = resample_for_autovot(file_path, tmpdirname)
+
+            with open(wav_filenames, 'w') as f:
+                f.write("{}\n".format(file_path))
+
+            with open(textgrid_filenames, 'w') as f:
+                f.write("{}\n".format(grid_path))
+
+            grid.write(grid_path)
+
+            if self.debug:
+                grid.write('/tmp/textgrid_from_conch.csv')
+                with open('/tmp/alt_wordlist.txt', 'w') as f:
+                    f.write("{}\n".format('/tmp/textgrid_from_conch.csv'))
+                subprocess.run(["auto_vot_decode.py", wav_filenames, '/tmp/alt_wordlist.txt', self.classifier_to_use, '--vot_tier', 'vot', '--vot_mark', 'vot', "--min_vot_length", str(self.min_vot_length), "--max_vot_length", str(self.max_vot_length), "--window_max", str(self.window_max), "--window_min", str(self.window_min)])
+            subprocess.run(["auto_vot_decode.py", wav_filenames, textgrid_filenames, self.classifier_to_use, '--vot_tier', 'vot', '--vot_mark', 'vot', '--csv_file', csv_path, "--min_vot_length", str(self.min_vot_length), "--max_vot_length", str(self.max_vot_length), "--window_max", str(self.window_max), "--window_min", str(self.window_min)])
+
+            return_list = []
+            with open(csv_path, "r") as f: 
+                f.readline()
+                for l, (b, e, *extra_data) in zip(f, vot_marks):
+                    _, time, vot, confidence = l.split(',')
+                    if "neg 0\n" ==  confidence:
+                        confidence = 0
+                    return_list.append((float(time), float(vot), float(confidence), *extra_data))
+            return return_list
+
+class AutoVOTAnalysisFunction(BaseAnalysisFunction):
+    def __init__(self, classifier_to_use=None, min_vot_length=15, max_vot_length=250, window_max=30, window_min=30, debug=False, arguments=None):
+        super(AutoVOTAnalysisFunction, self).__init__()
+        self._function = MeasureVOTPretrained(classifier_to_use=classifier_to_use, min_vot_length=min_vot_length, max_vot_length=max_vot_length, window_max=window_max, window_min=window_min, debug=debug)
+        self.requires_file = True
+        self.uses_segments = True
+        self.requires_segment_as_arg = True

conch/analysis/functions.py

@@ -15,6 +15,7 @@ class BaseAnalysisFunction(object):
     def __init__(self):
         self._function = print
         self.requires_file = False
+        self.requires_segment_as_arg = False
         self.uses_segments = False
         self.arguments = []
 
@@ -39,6 +40,8 @@ def __call__(self, segment):
         elif isinstance(segment, str) and not self.requires_file:
             signal, sr = librosa.load(safe_path(segment))
             return self._function(signal, sr, *self.arguments)
+        elif isinstance(segment, FileSegment) and self.requires_segment_as_arg:
+            return self._function(segment, *self.arguments)
         elif isinstance(segment, FileSegment) and self.requires_file and not self.uses_segments:
             beg, end = segment.begin, segment.end
             padding = segment['padding']

conch/analysis/segments.py

@@ -44,6 +44,8 @@ def __eq__(self, other):
             return False
         if self.channel != other.channel:
             return False
+        if self.properties != other.properties:
+            return False
         return True
 
     def __lt__(self, other):

tests/conftest.py

@@ -37,6 +37,9 @@ def praat_script_test_dir(test_dir):
 def soundfiles_dir(test_dir):
     return os.path.join(test_dir, 'soundfiles')
 
+@pytest.fixture(scope='session')
+def autovot_dir(test_dir):
+    return os.path.join(test_dir, 'autovot')
 
 @pytest.fixture(scope='session')
 def tts_dir(test_dir):
@@ -87,6 +90,29 @@ def base_filenames(soundfiles_dir):
                  if x.endswith('.wav')]
     return filenames
 
+@pytest.fixture(scope='session')
+def autovot_filenames(autovot_dir):
+    filenames = [os.path.join(autovot_dir, x)
+                 for x in os.listdir(autovot_dir)
+                 if x.endswith('.wav')]
+    return filenames
+
+@pytest.fixture(scope='session')
+def autovot_markings(test_dir, autovot_dir):
+    vot_markings = {}
+    with open(os.path.join(test_dir, "vot_marks"), "r") as f:
+        for x in f:
+            k = os.path.join(autovot_dir, x.split(' ')[0])
+            l = x.split(' ')[1:]
+            v = []
+            while l:
+                v.append((float(l.pop(0)), float(l.pop(0))))
+            vot_markings[k] = v
+    return vot_markings
+
+@pytest.fixture(scope='session')
+def autovot_correct_times():
+    return [(0.82, 0.005, 42.0803), (0.92, 0.005, 50.5166), (0.8, 0.005, 125.03), (0.76, 0.007, 93.6872), (0.756, 0.01, 101.213), (0.78, 0.009, 108.41), (0.774, 0.005, -62.3751), (0.807, 0.016, 85.4072), (0.64, 0.008, 70.7911), (0.753, 0.005, 63.7703), (0.692, 0.005, 103.689), (0.919, 0.021, 55.0379), (0.554, 0.005, -58.1071), (0.699, 0.024, 73.1509), (0.65, 0.011, 84.4646), (0.713, 0.022, 78.342), (0.66, 0.005, 50.3185), (0.67, 0.005, 67.515), (0.73, 0.005, 76.6631), (0.973, 0.005, -62.2706), (0.88, 0.009, 89.3114), (1.044, 0.005, 108.472), (0.704, 0.005, 93.9944), (0.93, 0.005, 65.8704), (0.73, 0.005, 111.866), (0.861, 0.005, 94.9302)]
 
 @pytest.fixture(scope='session')
 def praatpath():
@@ -114,7 +140,6 @@ def formants_func():
                                 window_length=0.025)
     return func
 
-
 @pytest.fixture(scope='session')
 def pitch_func():
     func = PitchTrackFunction(min_pitch=50, max_pitch=500, time_step=0.01)

tests/data/autovot

@@ -0,0 +1 @@
+../../autovot/experiments/data/tutorialExample/test/voiced/

tests/data/vot_marks

@@ -0,0 +1,26 @@
+cas7D_1054_24_3.wav 0.81000 0.93000
+cas7D_1054_26_1.wav 0.79000 0.87000
+cas7D_1054_26_2.wav 0.85000 0.89000
+cas7D_1054_26_3.wav 0.68000 0.79000
+cas7D_1054_28_1.wav 0.83000 0.92000
+cas7D_1054_28_2.wav 0.95000 0.98000
+cas7D_1054_28_3.wav 0.74000 0.84000
+cas7D_1054_30_1.wav 0.80000 0.87000
+cas7D_1054_30_2.wav 0.85000 0.95000
+cas7D_1054_30_3.wav 0.67000 0.74000
+cas7D_1054_32_1.wav 0.78000 0.86000
+cas7D_1054_32_2.wav 0.72000 0.82000
+cas7D_1054_32_3.wav 0.58000 0.67000
+cas7D_1054_34_1.wav 0.66000 0.74000
+cas7D_1054_34_2.wav 0.59000 0.68000
+cas7D_1054_34_3.wav 0.66000 0.75000
+cas7D_1054_36_1.wav 0.70000 0.80000
+cas7D_1054_36_2.wav 0.69000 0.80000
+cas7D_1054_36_3.wav 0.76000 0.86000
+cas7D_1144_32_3.wav 1.00000 1.10000
+cas7D_1144_34_1.wav 1.07000 1.11000
+cas7D_1144_34_2.wav 0.91000 0.98000
+cas7D_1144_34_3.wav 0.96000 1.03000
+cas7D_1144_36_1.wav 0.73000 0.80000
+cas7D_1144_36_2.wav 0.76000 0.86000
+cas7D_1144_36_3.wav 0.89000 0.98000

tests/test_analysis_autovot.py

@@ -0,0 +1,22 @@
+from conch.analysis.autovot import AutoVOTAnalysisFunction
+import librosa
+from statistics import mean
+import wave
+import pytest
+from conch.analysis.segments import SegmentMapping
+from conch import analyze_segments
+
+
+def test_autovot(autovot_filenames, autovot_markings, autovot_correct_times):
+    mapping = SegmentMapping()
+    for filename in autovot_filenames:
+        with wave.open(filename,'r') as f: 
+            length = f.getnframes() / float(f.getframerate())
+        mapping.add_file_segment(filename, 0, length, channel=0, vot_marks=autovot_markings[filename])
+    func = AutoVOTAnalysisFunction(classifier_to_use="autovot/experiments/models/bb_jasa.classifier", window_min=-30, window_max=30, min_vot_length=5, max_vot_length=100)
+    output = analyze_segments(mapping, func, multiprocessing=False)
+    output_l = sorted([output[x][0] for x in output], key=lambda x:x[0])
+    autovot_correct_times.sort(key=lambda x:x[0])
+    for o, truth in zip(output_l, autovot_correct_times):
+        assert o == truth
+