use torchaudio instead of librosa

data/get_sounds.py

@@ -51,7 +51,7 @@ def getBaseURL(soundscape = False, species = False,
         elif sort_by == 'title' :
             sort_by = 't'
         else :
-            raise NotImplementedError("sort_by cannot be set to " + sort_by)
+            raise NotImplementedError("sort_by cannot be set to " + str(sort_by))
         filters.append(SORT_BY_PATTERN.format(sort_by = sort_by))
     if since :
         if since == 'day' :
@@ -63,7 +63,7 @@ def getBaseURL(soundscape = False, species = False,
         elif since == 'year' :
             since = 'y'
         else :
-            raise NotImplementedError("since cannot be set to " + since)
+            raise NotImplementedError("since cannot be set to " + str(since))
         filters.append(SINCE_PATTERN.format(since = since))
 
     url = URL_SEARCH_PATTERN.format(filters = ','.join(filters))
@@ -155,13 +155,13 @@ def getSoundMetadata(html) :
     raise NotImplementedError('todo')
 
 
-def getRecordings(directory = '.',
-                  soundscape = False, species = False,
-                  validated = False, not_validated = False,
-                  sort_by = None,
-                  since = None,
-                  overwrite = False) :
-    """Download all the recordings corresponding to specific filters.
+def getSounds(directory = '.',
+              soundscape = False, species = False,
+              validated = False, not_validated = False,
+              sort_by = None,
+              since = None,
+              overwrite = False) :
+    """Download all the sounds corresponding to specific filters.
 
     :param directory: the path to save the recordings to.
     :param soundscape: Whether to get ambient sounds. If ambiant and species are False, 
@@ -200,4 +200,4 @@ def getRecordings(directory = '.',
     return None
 
 if __name__ == '__main__' :
-    getRecordings(directory = 'sounds', soundscape = True, validated = True)
+    getSounds(directory = 'sounds', soundscape = True, validated = True)

data/get_spectrograms.py

 # -*- coding: utf-8 -*-
 
-import librosa
 import os.path
-import pickle
-import warnings
+import torch
+import torchaudio
+import torchaudio.transforms as transforms
 
-def soundToSpectrogram(path, samplerate = 16000) :
-    """Return the Mel Spectrogram of an audio file.
+def soundToSpectrogram(path, sample_rate = 16000, duration = 30) :
+    r"""Return the Mel Spectrogram of an audio file.
 
-    :param filename: the path to the file.
-    :param samplerate: the sample rate to use when getting the spectrogram."""
+    Args :
+        path (str): the path to the file.
+        sample_rate (int, optional) : the sample rate in Hz to use when getting the spectrogram. Default : 16000 Hz
+        duration (int, optional) : the number of seconds the recording should last. Default : 30s
+    Returns :
+        torch.Tensor : Mel frequency spectrogram of size (2, 128, time)"""
+
+    # Load the recording
+    waveform, orig_freq = torchaudio.load(path)
+
+    # Convert to mono
+    waveform[:] = waveform.mean(dim = 0)
+
+    # Resample
+    resample = transforms.Resample(orig_freq = orig_freq, new_freq = sample_rate)
+    waveform = resample(waveform)
+
+    # Extract a certain duration
+    try :
+        waveform = waveform[:, :duration * sample_rate]
+    except :
+        raise TypeError('the recording is too long or too short')
+
+    # Compute the Mel-spectrogram
+    melspectrogram = transforms.MelSpectrogram(sample_rate = sample_rate)
+    spectrogram = melspectrogram(waveform)
 
-    with warnings.catch_warnings() :
-        warnings.filterwarnings('ignore', category = UserWarning)
-        signal, samplerate = librosa.load(path, sr = samplerate)
-        spectrogram = librosa.feature.melspectrogram(y = signal, sr = samplerate)
     return spectrogram
 
 
-def getSpectrograms(source_directory, target_directory, samplerate = 16000, overwrite = False) :
-    """Compute the Mel spectrograms of all the sound files.
+def getSpectrograms(dir_source, dir_target,
+                    sample_rate = 16000,
+                    overwrite = False,
+                    duration = 30) :
+    r"""Compute the Mel spectrograms of all the sound files and write them to memory.
 
-    :param source_path: the directory where the sound files are.
-    :param target_path: the directory to save the the spectrograms to.
-    :param samplerate: the sample rate to use when getting the spectrogram."""
+    Args :
+        dir_source (str) : the directory where the sound files are. 
+        dir_target (str) : the directory to save the the spectrograms to.
+        sample_rate (int, optional) : the sample rate in Hz to use when getting the spectrogram. Default : 16000 Hz
+        overwrite (bool, optional) : whether to compute all spectrograms again. Default : False
+        duration (int, optional) : the number of seconds the recording should last. Default : 30s"""
 
-    for filename in os.listdir(source_directory) :
+    for filename in os.listdir(dir_source) :
         title, _ = os.path.splitext(filename)
-        source_path = os.path.join(source_directory, filename)
-        target_path = os.path.join(target_directory, title + '.bin')
-        
-        if overwrite :
-            mode = 'w+b'
-        else :
-            mode = 'x+b'
-
-        print('Computing the Mel spectrogram of {filename} ...'.format(filename = filename))
-        try :
-            with open(target_path, mode) as file :
-                spectrogram = soundToSpectrogram(source_path, samplerate = samplerate)
-                spectrogram.tofile(file)
-            print('Success.')
-        except FileExistsError as exception :
-            print('Failure : {exception}'.format(exception = exception))
+        path_source = os.path.join(dir_source, filename)
+        path_target = os.path.join(dir_target, title + '.pt')
 
+        print('Computing the Mel-spectrogram of {filename} ...'.format(filename = filename))
+        if overwrite or not(os.path.isfile(path_target)) :
+            try :
+                spectrogram = soundToSpectrogram(path_source, sample_rate = sample_rate, duration = duration)
+                torch.save(spectrogram, path_target)
+                print('Success.')
+            except Exception as exception :
+                print('Failure : {exception}.'.format(exception = exception))
+        print('Mel-spectrogram already computed.')
+            
 if __name__ == '__main__' :
     getSpectrograms('sounds', 'spectrograms')

data/requirements.txt

 bs4
-librosa
+torch
+torchaudio