add audio read plugin for EPG Systems' AQ8 files

Author: bjarthur

src/audio-read-plugin.py

@@ -1,8 +1,21 @@
-# a function that inputs the full path to a file containing the audio recording,
-# an interval of time, and some keyword arguments and returns the sampling
-# rate, shape of entire recording (not just the interval), and requested data as int16
+#a function that inputs the full path (including possibly a recording letter) to
+#a file containing the audio recording, an interval of time, and some keyword
+#arguments and returns the sampling rate, shape of entire recording (not just
+#the interval), and requested data as int16.  if {start,stop}_tic are None, return
+#the entire recording
 def audio_read(fullpath, start_tic, stop_tic, **kw):
 
     # load data, determine sampling rate and length, and do any special processing
 
     return sampling_rate, nsamples_nchannels, slice_of_data
+
+# a function that returns a list of file extensions which this plugin can handle
+def audio_read_exts(**kw):
+    return []  # e.g. ['.wav', '.WAV']
+
+# a function that returns a dictionary that maps logical recordings to channels in the file
+def audio_read_rec2ch(**kw):
+    return {}  # e.g. {'A':[0], 'B':[1]}, or {'A':[0,1]}
+
+def audio_read_init(**kw):
+    pass

src/classify

@@ -78,6 +78,8 @@ def main():
   def audio_read(wav_path, start_tic=None, stop_tic=None):
       return audio_read_module.audio_read(wav_path, start_tic, stop_tic,
                                           **FLAGS.audio_read_plugin_kwargs)
+  def audio_read_rec2ch():
+      return audio_read_module.audio_read_rec2ch(**FLAGS.audio_read_plugin_kwargs)
 
   sys.path.append(os.path.dirname(FLAGS.video_read_plugin))
   video_read_module = importlib.import_module(os.path.basename(FLAGS.video_read_plugin))
@@ -244,7 +246,13 @@ def main():
     else:
       adjusted_probability = probability_matrix[:,ch]
     waveform = adjusted_probability*np.iinfo(np.int16).max
-    filename = os.path.splitext(FLAGS.wav)[0]+'-'+labels[ch]+'.wav'
+    if len(audio_read_rec2ch()) == 1:
+      withoutext = os.path.splitext(FLAGS.wav)[0]
+    else:
+      tmp = FLAGS.wav.split('-')
+      withext, rec = '-'.join(tmp[:-1]), tmp[-1]
+      withoutext = os.path.splitext(withext)[0]+'-'+rec
+    filename = withoutext+'-'+labels[ch]+'.wav'
     wavfile.write(filename, int(sample_rate), waveform.astype('int16'))
 
 if __name__ == '__main__':

src/congruence

@@ -11,7 +11,8 @@
 #     --measure=label \
 #     --nprobabilities=20 \
 #     --audio_tic_rate=2500 \
-#     --parallelize=1
+#     --parallelize=1 \
+#     --has_rec=False
 
 import argparse
 import sys
@@ -164,7 +165,12 @@ def main():
   thresholds=set()
   for wavdir in wavdirs:
     for wavfile in wavdirs[wavdir]:
-      wavfile_noext = os.path.splitext(wavfile)[0]
+      if FLAGS.has_rec == "False":
+        wavfile_noext  = os.path.splitext(wavfile)[0]
+      else:
+        tmp = wavfile.split('-')
+        withext, rec = '-'.join(tmp[:-1]), tmp[-1]
+        wavfile_noext  = os.path.splitext(withext)[0]+'-'+rec
 
       if not labels:
         with open(os.path.join(FLAGS.basepath,wavdir,wavfile_noext+'-ethogram.log'), 'r') as fid:
@@ -704,6 +710,10 @@ if __name__ == "__main__":
       '--parallelize',
       default=0,
       type=int)
+  parser.add_argument(
+      '--has_rec',
+      default='False',
+      type=str)
 
   FLAGS, unparsed = parser.parse_known_args()
 

src/data.py

@@ -118,12 +118,15 @@ def __init__(self, data_dir,
         self.np_rng = np.random.default_rng(None if random_seed_batch==-1 else random_seed_batch)
 
         sys.path.append(os.path.dirname(audio_read_plugin))
-        self.audio_read_plugin = os.path.basename(audio_read_plugin)
+        audio_read_plugin = os.path.basename(audio_read_plugin)
         self.audio_read_plugin_kwargs = audio_read_plugin_kwargs
+        self.audio_read_module = importlib.import_module(audio_read_plugin)
+        self.audio_read_module.audio_read_init(**self.audio_read_plugin_kwargs)
 
         sys.path.append(os.path.dirname(video_read_plugin))
-        self.video_read_plugin = os.path.basename(video_read_plugin)
+        video_read_plugin = os.path.basename(video_read_plugin)
         self.video_read_plugin_kwargs = video_read_plugin_kwargs
+        self.video_read_module = importlib.import_module(video_read_plugin)
 
         self.prepare_data_index(shiftby,
                                 labels_touse, kinds_touse,
@@ -139,14 +142,12 @@ def __init__(self, data_dir,
         signal.signal(signal.SIGTERM, term)
 
     def audio_read(self, fullpath, start_tic=None, stop_tic=None):
-        audio_read_module = importlib.import_module(self.audio_read_plugin)
-        return audio_read_module.audio_read(fullpath, start_tic, stop_tic,
-                                            **self.audio_read_plugin_kwargs)
+        return self.audio_read_module.audio_read(fullpath, start_tic, stop_tic,
+                                                 **self.audio_read_plugin_kwargs)
 
     def video_read(self, fullpath, start_frame=None, stop_frame=None):
-        video_read_module = importlib.import_module(self.video_read_plugin)
-        return video_read_module.video_read(fullpath, start_frame, stop_frame,
-                                            **self.video_read_plugin_kwargs)
+        return self.video_read_module.video_read(fullpath, start_frame, stop_frame,
+                                                 **self.video_read_plugin_kwargs)
 
     def catalog_overlaps(self, data):
         data.sort(key=lambda x: x['ticks'][0])

src/ethogram

@@ -2,10 +2,10 @@
 
 # apply per-class thresholds to discretize probabilities
 
-# ethogram <logdir> <model> <thresholds-file> <wav-file> <wav-tic-rate>
+# ethogram <logdir> <model> <thresholds-file> <wav-file> <wav-tic-rate> <has_rec>
 
 # e.g.
-# ethogram `pwd`/trained-classifier 1k 50 `pwd`/groundtruth-data/round1/20161207T102314_ch1_p1.wav 5000
+# ethogram `pwd`/trained-classifier 1k 50 `pwd`/groundtruth-data/round1/20161207T102314_ch1_p1.wav 5000 False
 
 import sys
 import os
@@ -29,19 +29,28 @@ print("hostname = "+socket.gethostname())
 
 try:
 
-  _,logdir,model,thresholds_file,wav_file,audio_tic_rate = argv
+  _,logdir,model,thresholds_file,wav_file,audio_tic_rate,has_rec = argv
   print('logdir: '+logdir)
   print('model: '+model)
   print('thresholds_file: '+thresholds_file)
   print('wav_file: '+wav_file)
   print('audio_tic_rate: '+audio_tic_rate)
+  print('has_rec: '+has_rec)
   audio_tic_rate=float(audio_tic_rate)
 
-  if not os.path.isfile(wav_file):
-    print('cannot find WAV file')
-    exit()
   wavpath, wavname = os.path.split(wav_file)
-  wavname_noext = os.path.splitext(wavname)[0]
+  if has_rec == "False":
+    if not os.path.isfile(wav_file):
+      print('cannot find WAV file')
+      exit()
+    wavname_noext  = os.path.splitext(wavname)[0]
+  else:
+    tmp = wavname.split('-')
+    withext, rec = '-'.join(tmp[:-1]), tmp[-1]
+    wavname_noext  = os.path.splitext(withext)[0]+'-'+rec
+    if not os.path.isfile(os.path.join(wavpath, withext)):
+      print('cannot find WAV file')
+      exit()
 
   precision_recall_ratios, thresholds = read_thresholds(logdir, model, thresholds_file)
 

src/gui/controller.py

@@ -1043,15 +1043,17 @@ def _validation_test_files(files_string, comma=True):
             return [','.join(wavfiles)] if comma else list(wavfiles)
     elif os.path.dirname(files_string.rstrip(os.sep)) == V.groundtruth_folder.value.rstrip(os.sep):
         dfs = []
-        for csvfile in filter(lambda x: x.endswith('.csv'), os.listdir(files_string)):
+        for csvfile in filter(lambda x: os.path.splitext(x)[1] in M.audio_read_exts(),
+                              os.listdir(files_string)):
             filepath = os.path.join(files_string, csvfile)
             if os.path.getsize(filepath) > 0:
                 dfs.append(pd.read_csv(filepath, header=None, index_col=False))
         if dfs:
             df = pd.concat(dfs)
             wavfiles = sorted(list(set(df.loc[df[3]=="annotated"][0])))
             return [','.join(wavfiles)] if comma else list(wavfiles)
-    elif files_string.lower().endswith('.wav'):
+    elif os.path.splitext(files_string[:-2 if len(M.audio_read_rec2ch())>1 else None])[1] \
+                in M.audio_read_exts():
         return [files_string] if comma else files_string.split(',')
     elif files_string!='':
         with open(files_string, "r") as fid:
@@ -1827,7 +1829,7 @@ def classify_succeeded(modeldir, wavfile, reftime):
     with open(os.path.join(modeldir, 'labels.txt'), 'r') as fid:
         labels = fid.read().splitlines()
     for x in labels:
-        if not recent_file_exists(wavfile[:-4]+'-'+x+'.wav', reftime, True):
+        if not recent_file_exists(M.trim_ext(wavfile)+'-'+x+'.wav', reftime, True):
             return False
     return True
 
@@ -1839,7 +1841,7 @@ async def _classify_actuate(wavfiles):
     wavfile = wavfiles.pop(0)
     currtime = time.time()
     logdir, model, _, check_point = M.parse_model_file(V.model_file.value)
-    logfile = os.path.splitext(wavfile)[0]+'-classify.log'
+    logfile = M.trim_ext(wavfile)+'-classify.log'
     args = ["--context="+V.context.value,
             "--shiftby="+V.shiftby.value,
             "--loss="+V.loss.value,
@@ -1900,7 +1902,7 @@ def ethogram_succeeded(modeldir, ckpt, wavfile, reftime):
         row1 = next(csvreader)
     precision_recalls = row1[1:]
     for x in precision_recalls:
-        if not recent_file_exists(wavfile[:-4]+'-predicted-'+x+'pr.csv', reftime, True):
+        if not recent_file_exists(M.trim_ext(wavfile)+'-predicted-'+x+'pr.csv', reftime, True):
             return False
     return True
 
@@ -1918,14 +1920,15 @@ async def _ethogram_actuate(i, wavfiles, threads, results):
         thresholds_file = os.path.basename(V.model_file.value)
     else:
         thresholds_file = 'thresholds.ckpt-'+check_point+'.csv'
-    logfile = os.path.splitext(wavfile)[0]+'-ethogram.log'
+    logfile = M.trim_ext(wavfile)+'-ethogram.log'
     jobid = generic_actuate("ethogram", logfile, M.ethogram_where,
                             M.ethogram_ncpu_cores,
                             M.ethogram_ngpu_cards,
                             M.ethogram_ngigabytes_memory,
                             M.ethogram_cluster_flags,
                             logdir, model, thresholds_file, wavfile,
-                            str(M.audio_tic_rate))
+                            str(M.audio_tic_rate),
+                            "False" if len(M.audio_read_rec2ch()) == 1 else "True")
     displaystring = "ETHOGRAM "+os.path.basename(wavfile)
     if jobid:
         displaystring += " ("+jobid+")"
@@ -2048,7 +2051,8 @@ async def congruence_actuate():
                             "--measure="+V.congruence_measure.value,
                             "--nprobabilities="+str(M.nprobabilities),
                             "--audio_tic_rate="+str(M.audio_tic_rate),
-                            "--parallelize="+str(M.congruence_parallelize))
+                            "--parallelize="+str(M.congruence_parallelize),
+                            "--has_rec="+("False" if len(M.audio_read_rec2ch()) == 1 else "True"))
     displaystring = "CONGRUENCE "+os.path.basename(all_files[0])
     if jobid:
         displaystring += " ("+jobid+")"
@@ -2098,12 +2102,18 @@ def wavcsv_files_callback():
     if len(V.file_dialog_source.selected.indices)==0:
         bokehlog.info('ERROR: a file(s) must be selected in the file browser')
         return
-    filename = V.file_dialog_source.data['names'][V.file_dialog_source.selected.indices[0]]
-    files = os.path.join(M.file_dialog_root, filename)
-    for i in range(1, len(V.file_dialog_source.selected.indices)):
+    files = []
+    for i in range(len(V.file_dialog_source.selected.indices)):
         filename = V.file_dialog_source.data['names'][V.file_dialog_source.selected.indices[i]]
-        files += ','+os.path.join(M.file_dialog_root, filename)
-    V.wavcsv_files.value = files
+        if os.path.splitext(filename)[1] in M.audio_read_exts():
+            if len(M.audio_read_rec2ch()) == 1:
+                files.append(os.path.join(M.file_dialog_root, filename))
+            else:
+                files.extend([os.path.join(M.file_dialog_root, filename)+'-'+k
+                              for k in M.audio_read_rec2ch().keys()])
+        else:
+            files.append(os.path.join(M.file_dialog_root, filename))
+    V.wavcsv_files.value = ','.join(files)
 
 def groundtruth_callback():
     if len(V.file_dialog_source.selected.indices)>=2:
@@ -2123,8 +2133,12 @@ def _validation_test_files_callback():
       filename = V.file_dialog_source.data['names'][V.file_dialog_source.selected.indices[0]]
       filepath = os.path.join(M.file_dialog_root, filename)
     if nindices<2:
-        if filepath.lower().endswith('.wav'):
-            return os.path.basename(filepath)
+        if os.path.splitext(filepath)[1] in M.audio_read_exts():
+            if len(M.audio_read_rec2ch()) == 1:
+                return os.path.basename(filepath)
+            else:
+                return ','.join([os.path.basename(filepath)+'-'+k
+                                 for k in M.audio_read_rec2ch().keys()])
         else:
             return filepath
     else:

src/gui/model.py

@@ -78,14 +78,27 @@ def isused(sound):
     return np.where([x['file']==sound['file'] and x['ticks']==sound['ticks'] \
                      for x in used_sounds])[0]
 
+def trim_ext(wavfile):
+    if len(audio_read_rec2ch()) == 1:
+        withoutext = os.path.splitext(wavfile)[0]
+    else:
+        tmp = wavfile.split('-')
+        withext, rec = '-'.join(tmp[:-1]), tmp[-1]
+        withoutext = os.path.splitext(withext)[0]+'-'+rec
+    return withoutext
+
 def save_annotations():
     global nrecent_annotations
     if nrecent_annotations>0:
         fids = {}
         csvwriters = {}
         csvfiles_current = set([])
-        for wavfile in set([os.path.join(*x['file']) for x in annotated_sounds if x["label"]!=""]):
-            csvfile = wavfile[:-4]+"-annotated-"+songexplorer_starttime+".csv"
+        wavfiles = set()
+        for sound in annotated_sounds:
+            if not sound["label"]:  continue
+            wavfiles |= set([trim_ext(os.path.join(*sound["file"]))])
+        for wavfile in wavfiles:
+            csvfile = wavfile+"-annotated-"+songexplorer_starttime+".csv"
             annotated_csvfiles_all.add(csvfile)
             csvfiles_current.add(csvfile)
             fids[wavfile] = open(os.path.join(V.groundtruth_folder.value, csvfile),
@@ -98,10 +111,11 @@ def save_annotations():
         corrected_sounds=[]
         for annotation in annotated_sounds:
             if annotation['label']!="" and not annotation['label'].isspace():
-                csvwriters[os.path.join(*annotation['file'])].writerow(
+                wavfile_noext = trim_ext(os.path.join(*annotation['file']))
+                csvwriters[wavfile_noext].writerow(
                         [annotation['file'][1],
-                        annotation['ticks'][0], annotation['ticks'][1],
-                        'annotated', annotation['label']])
+                         annotation['ticks'][0], annotation['ticks'][1],
+                         'annotated', annotation['label']])
             iused = isused(annotation)
             if len(iused)>0 and used_sounds[iused[0]]['kind']=='annotated':
                 corrected_sounds.append(annotation)
@@ -110,10 +124,14 @@ def save_annotations():
                                           x['ticks'][1], 'annotated', x['label']] \
                                          for x in corrected_sounds], \
                                         columns=['file','start','stop','kind','label'])
-            for wavfile in set([os.path.join(*x['file']) for x in corrected_sounds]):
+            wavfiles = set()
+            for sound in corrected_sounds:
+                wavfile_noext = trim_ext(os.path.join(*sound["file"]))
+                wavfiles |= set([wavfile_noext])
+            for wavfile in wavfiles:
                 wavdir, wavbase = os.path.split(wavfile)
                 wavpath = os.path.join(V.groundtruth_folder.value, wavdir)
-                for csvbase in filter(lambda x: x.startswith(wavbase[:-4]) and
+                for csvbase in filter(lambda x: x.startswith(os.path.splitext(wavbase)[0]) and
                                                 x.endswith(".csv") and
                                                 "-annotated" in x and
                                                 songexplorer_starttime not in x,
@@ -235,7 +253,8 @@ def init(_bokeh_document, _configuration_file, _use_aitch):
     global context_width_sec0, context_offset_sec0
     global xcluster, ycluster, zcluster, ndcluster, tic2pix_max, snippet_width_pix, ilayer, ispecies, iword, inohyphen, ikind, nlayers, layers, species, words, nohyphens, kinds, used_labels, snippets_gap_sec, snippets_tic, snippets_gap_tic, snippets_decimate_by, snippets_pix, snippets_gap_pix, context_decimate_by, context_width_tic, context_offset_tic, context_sound, isnippet, xsnippet, ysnippet, file_nframes, context_midpoint_tic, ilabel, used_sounds, used_starts_sorted, used_stops, iused_stops_sorted, annotated_sounds, annotated_starts_sorted, annotated_stops, iannotated_stops_sorted, annotated_csvfiles_all, nrecent_annotations, clustered_sounds, clustered_activations, used_recording2firstsound, clustered_starts_sorted, clustered_stops, iclustered_stops_sorted, songexplorer_starttime, history_stack, history_idx, wizard, action, function, statepath, state, file_dialog_root, file_dialog_filter, nearest_sounds, status_ticker_queue, waitfor_job, dfs, remaining_isounds
     global user_changed_recording, user_copied_parameters
-    global audio_read, video_read, detect_labels, doubleclick_annotation, context_data, context_data_istart, model, video_findfile
+    global audio_read, audio_read_exts, audio_read_rec2ch
+    global video_read, detect_labels, doubleclick_annotation, context_data, context_data_istart, model, video_findfile
     global detect_parameters, doubleclick_parameters, model_parameters, cluster_parameters
 
     bokeh_document = _bokeh_document
@@ -253,9 +272,11 @@ def init(_bokeh_document, _configuration_file, _use_aitch):
 
     sys.path.insert(0,os.path.dirname(audio_read_plugin))
     audio_read_module = importlib.import_module(os.path.basename(audio_read_plugin))
+    audio_read_module.audio_read_init(**audio_read_plugin_kwargs)
     def audio_read(wav_path, start_tic=None, stop_tic=None):
-        return audio_read_module.audio_read(wav_path, start_tic, stop_tic,
-                                            **audio_read_plugin_kwargs)
+        return audio_read_module.audio_read(wav_path, start_tic, stop_tic, **audio_read_plugin_kwargs)
+    def audio_read_exts(): return audio_read_module.audio_read_exts(**audio_read_plugin_kwargs)
+    def audio_read_rec2ch(): return audio_read_module.audio_read_rec2ch(**audio_read_plugin_kwargs)
 
     sys.path.insert(0,os.path.dirname(video_read_plugin))
     video_read_module = importlib.import_module(os.path.basename(video_read_plugin))

src/highpass-filter.py

@@ -33,3 +33,12 @@ def audio_read(wav_path, start_tic, stop_tic, cutoff=1, order=2):
     data_unpadded = data_filtered[padlenL:-padlenR or None, :]
 
     return sampling_rate, data.shape, data_unpadded
+
+def audio_read_exts(**kw):
+    return ['.wav', '.WAV']
+
+def audio_read_rec2ch(**kw):
+    return {'A':[0]}
+
+def audio_read_init(**kw):
+    pass