Minor cleanup including whitespace.

README.md

@@ -145,5 +145,3 @@ Scaling
 The fingerprint database records 2^20 (~1M) distinct fingerprints, with (by default) 100 entries for each fingerprint bucket.  When the bucket fills, track entries are dropped at random; since matching depends only on making a minimum number of matches, but no particular match, dropping some of the more popular ones does not prevent matching.  The Matlab version has been successfully used for databases of 100k+ tracks.  Reducing the hash density (`--density`) leads to smaller reference database size, and the capacity to record more reference items before buckets begin to fill; a density of 7.0 works well.
 
 Times (in units of 256 samples, i.e., 23 ms at the default 11kHz sampling rate) are stored in the bottom 14 bits of each database entry, meaning that times larger than 2^14*0.023 = 380 sec, or about 6 mins, are aliased.  If you want to correctly identify time offsets in tracks longer than this, you need to use a larger `--maxtimebits`; e.g. `--maxtimebits 16` increases the time range to 65,536 frames, or about 25 minutes at 11 kHz.  The trade-off is that the remaining bits in each 32 bit entry (i.e., 18 bits for the default 14 bit times) are used to store the track ID.  Thus, by default, the database can only remember 2^18 = 262k tracks; using a larger `--maxtimebits` will reduce this; similarly, you can increase the number of distinct tracks by reducing `--maxtimebits`, which doesn't prevent matching tracks, but progressively reduces discrimination as the number of distinct time slots reduces (and can make the reported time offsets, and time ranges for `--find-time-ranges`, completely wrong for longer tracks).
-
-

audfprint.py

@@ -10,17 +10,23 @@
 """
 from __future__ import division, print_function
 
-import multiprocessing  # for new/add
-import os  # For command line interface
-import sys  # For __main__
-import time  # For reporting progress time
-
-import docopt  # For command line interface
-import joblib  # for match
-
-import audfprint_analyze  # The actual analyzer class/code
-import audfprint_match  # Access to match functions, used in command line interface
-import hash_table  # My hash_table implementation
+# For reporting progress time
+import time
+# For command line interface
+import docopt
+import os
+# For __main__
+import sys
+# For multiprocessing options
+import multiprocessing
+import joblib
+
+# The actual analyzer class/code
+import audfprint_analyze
+# Access to match functions, used in command line interface
+import audfprint_match
+# My hash_table implementation
+import hash_table
 
 
 if sys.version_info[0] >= 3:
@@ -413,13 +419,13 @@ def main(argv):
 
     # Setup the analyzer if we're using one (i.e., unless "merge")
     analyzer = setup_analyzer(args) if not (
-            cmd is "merge" or cmd is "newmerge"
-            or cmd is "list" or cmd is "remove") else None
+            cmd == "merge" or cmd == "newmerge"
+            or cmd == "list" or cmd == "remove") else None
 
     precomp_type = 'hashes'
 
     # Set up the hash table, if we're using one (i.e., unless "precompute")
-    if cmd is not "precompute":
+    if cmd != "precompute":
         # For everything other than precompute, we need a database name
         # Check we have one
         dbasename = args['--dbase']

audfprint_analyze.py

@@ -9,19 +9,23 @@
 
 from __future__ import division, print_function
 
-import glob  # For glob2hashtable, localtester
 import os
-import struct  # For reading/writing hashes to file
-import time  # For glob2hashtable, localtester
-
 import numpy as np
+
 import scipy.signal
 
+# For reading/writing hashes to file
+import struct
+
+# For glob2hashtable, localtester
+import glob
+import time
+
 import audio_read
-import hash_table  # For utility, glob2hashtable
+# For utility, glob2hashtable
+import hash_table
 import stft
 
-
 # ############### Globals ############### #
 # Special extension indicating precomputed fingerprint
 PRECOMPEXT = '.afpt'
@@ -80,6 +84,9 @@ def landmarks2hashes(landmarks):
     the three remaining values.
     """
     landmarks = np.array(landmarks)
+    # Deal with special case of empty landmarks.
+    if landmarks.shape[0] == 0:
+      return np.zeros((0, 2), dtype=np.int32)
     hashes = np.zeros((landmarks.shape[0], 2), dtype=np.int32)
     hashes[:, 0] = landmarks[:, 0]
     hashes[:, 1] = (((landmarks[:, 1] & B1_MASK) << B1_SHIFT)
@@ -182,7 +189,7 @@ def spreadpeaks(self, peaks, npoints=None, width=4.0, base=None):
             self.__sp_width = width
             self.__sp_len = npoints
             self.__sp_vals = np.exp(-0.5 * ((np.arange(-npoints, npoints + 1)
-                                             / width) ** 2))
+                                             / width)**2))
         # Now the actual function
         for pos, val in peaks:
             vec = np.maximum(vec, val * self.__sp_vals[np.arange(npoints)
@@ -195,7 +202,7 @@ def _decaying_threshold_fwd_prune(self, sgram, a_dec):
         """
         (srows, scols) = np.shape(sgram)
         sthresh = self.spreadpeaksinvector(
-                np.max(sgram[:, :np.minimum(10, scols)], axis=1), self.f_sd
+            np.max(sgram[:, :np.minimum(10, scols)], axis=1), self.f_sd
         )
         # Store sthresh at each column, for debug
         # thr = np.zeros((srows, scols))
@@ -399,7 +406,7 @@ def wavfile2hashes(self, filename):
                 query_hashes = []
                 for peaklist in peaklists:
                     query_hashes.append(landmarks2hashes(
-                            self.peaks2landmarks(peaklist)))
+                        self.peaks2landmarks(peaklist)))
                 query_hashes = np.concatenate(query_hashes)
             else:
                 query_hashes = landmarks2hashes(self.peaks2landmarks(peaks))

audfprint_match.py

@@ -21,10 +21,11 @@
 except:
     pass
 
-import audfprint_analyze  # for localtest and illustrate
+import audfprint_analyze
 import audio_read
 import stft
 
+
 def process_info():
     rss = usrtime = 0
     p = psutil.Process(os.getpid())

audio_read.py

@@ -39,7 +39,7 @@
 
 # If ffmpeg is unavailable, you can set HAVE_FFMPEG to False which will cause
 # soundfile reads to go via scipy.io.wavfile.  However, this means that only
-# *.wav files are supported *and* they must already be resampled to the 
+# *.wav files are supported *and* they must already be resampled to the
 # system sampling rate (e.g. 11025 Hz).
 
 HAVE_FFMPEG = True

stft.py

@@ -1,6 +1,6 @@
-"""Provide stft to avoid librosa dependency. 
+"""Provide stft to avoid librosa dependency.
 
-This implementation is based on routines from 
+This implementation is based on routines from
 https://github.com/tensorflow/models/blob/master/research/audioset/mel_features.py
 """
 
@@ -71,8 +71,8 @@ def stft(signal, n_fft, hop_length=None, window=None):
       values.  Defaults to n_fft.
 
   Returns:
-    2D np.array where each column contains the complex values of the 
-    fft_length/2+1 unique values of the FFT for the corresponding frame of 
+    2D np.array where each column contains the complex values of the
+    fft_length/2+1 unique values of the FFT for the corresponding frame of
     input samples ("spectrogram transposition").
   """
   if window is None: