.gitmodules

@@ -13,3 +13,6 @@
 [submodule "src/bodycontact"]
 	path = src/bodycontact
 	url = https://github.com/psychoinformatics-de/studyforrest-paper-bodycontactannotation.git
+[submodule "src/voice"]
+	path = src/voice
+	url = https://github.com/chrhaeusler/studyforrest-paper-speechannotation

code/importer/audio-description-by-word.py

@@ -0,0 +1,29 @@
+#!/usr/bin/python
+'''
+'''
+from os.path import join as opj
+import pandas as pd
+
+# read the annotation
+# TO DO: skiprows flagged with '???' here already?
+df = pd.read_csv(opj('src', 'voice', 'data', 'audio-description-by-word.csv'))
+
+# # drop rows with whole sentences as flagged with '???' in column 1
+df = df.loc[df.iloc[:,1] != '???']
+# reset index
+# df.index = range(0, len(df))
+
+# convert the cleaned columns to float64
+df.iloc[:, 0:2] = df.iloc[:, 0:2].astype('float64')
+
+# replace column 1 (end) with duration (end - start)
+df.iloc[:,1] = df.iloc[:,1] - df.iloc[:,0]
+
+# apply BIDS standard column names
+df.rename(columns=dict(start='onset', end='duration'), inplace=True)
+
+df.to_csv(
+    opj('researchcut', 'audio-description-by-word.tsv'),
+    index=False,
+    sep='\t',
+    float_format='%.3f')

code/importer/locations.py

@@ -17,5 +17,6 @@ df.insert(1, 'duration', shot_durations)
 
 df.to_csv(
     opj('researchcut', 'locations.tsv'),
+    index=False,
     sep='\t',
-    index=False)
+    float_format='%.3f')

code/importer/speech_vocalization.py

@@ -0,0 +1,48 @@
+#!/usr/bin/python
+'''
+'''
+from os.path import join as opj
+import pandas as pd
+
+
+def time_stamp_to_sec(t_stamp='01:50:34:01'):
+    '''
+    Input:
+        time stamp (str) in format HH:MM:SS:Frame
+
+    Output:
+        time point in seconds (float)
+    '''
+    splitted_stamp = t_stamp.split(':')
+    milliseconds = (int(splitted_stamp[0]) * 60 * 60 * 1000) +\
+                        (int(splitted_stamp[1]) * 60 * 1000) +\
+                        (int(splitted_stamp[2]) * 1000) +\
+                        (int(splitted_stamp[3]) * 40)
+
+    seconds = milliseconds / 1000.0
+
+    return seconds
+
+
+# read the annotation
+df = pd.read_csv(opj('src', 'voice', 'speech_vocalization.csv'))
+
+# filter for rows that contain an #-flag indicating missing timing
+df = df.loc[df.iloc[:, 0].str.contains('#') == False]
+df = df.loc[df.iloc[:, 1].str.contains('#') == False]
+
+# convert time stamps to
+df.iloc[:, 0] = df.iloc[:, 0].apply(time_stamp_to_sec)
+df.iloc[:, 1] = df.iloc[:, 1].apply(time_stamp_to_sec)
+
+# replace column 1 (end) with duration (end - start)
+df.iloc[:,1] = df.iloc[:,1] - df.iloc[:,0]
+
+# apply BIDS standard column names
+df.rename(columns=dict(start='onset', end='duration'), inplace=True)
+
+df.to_csv(
+    opj('researchcut', 'speech_vocalization.tsv'),
+    index=False,
+    sep='\t',
+    float_format='%.3f')

code/researchcut2segments.py

@@ -199,25 +199,14 @@ if __name__ == "__main__":
 
         # correct for the stimulus used to annotate the audiotrack
         if annotated_time == 'aomovie':
-            # the files
-            # forrestgump_researchcut_ad_ger.flac and
-            # german_dvd_5.1_48000hz_488kb_research_cut_aligned_cutted_narrator_muted_48000Hz.flac
-            # (that contain the audio description) were originally lagging
-            # behind for XYZ msec and were shiftet forward
-            # by one frame (40ms) in respect to the reference file
-            # forrestgump_researchcut_ger.mkv
-
-            # 1st, correct for shifting the narrator (incl. dialogue) 40ms
-            # to the front before annotating the narrator/dialogue
-            onset_in_seg += 0.040
-
-            # 2nd, correct for the offset between the (unshifted) audio
-            # description and the audiovisual movie
-            # -> the offset is varying +/- one frame (40 ms) around 0
+            # first, correct for the offset between the (unshifted) audio
+            # description and audiovisual movie
+            # it turned out the offset is varying +/- one frame (40 ms) around 0
+            # across the course of the whole stimuli
             onset_in_seg -= 0.000
 
-            # 3rd, correct for the offset between whole stimulus
-            # (audiovisual or audio-only) and its segments
+            # second, correct for the offset between whole stimulus
+            # (audiovisual or audio-description) and its segments
             if target_time == 'avmovie':
                 onset_in_seg = fix_audio_movie_segments(
                     AUDIO_AV_OFFSETS,
@@ -237,10 +226,8 @@ if __name__ == "__main__":
             # all splendid for now
             pass
 
-        else:
-            raise ValueError('%s is an unknown annotation', basename(input_file))
-
         row['onset'] = round(onset_in_seg, 3)
+        row['duration'] = round(row['duration'], 3)
 
         # append that shit
         run_events[run].append(row)