more upload

Author: ghunkins

Data Processing/generate.py

@@ -0,0 +1,56 @@
+"""
+Script for linking to MATLAB to generate
+an HRTF dataset.
+"""
+
+import matlab.engine
+import itertools
+import os
+import random
+import time
+
+root = "/Users/ghunk/Desktop/GRADUATE/CSC_464/Final_Project/Dataset/"
+
+paths = {"tools": root + "tools/LISTEN/",
+		 "timit": root + "voice_data/TIMIT_wav/TIMIT_all/",
+		 "hrir":  root + "hrtf_data/LISTEN_HRIR/",
+		 "save":  root + "binaural_random/"}
+
+elevations = [-45, -30, -15, 0, 15, 30, 45]
+azimuths = [15*x for x in range(24)]
+combos = list(itertools.product(elevations, azimuths))
+
+def main():
+	eng = matlab.engine.connect_matlab()
+	eng.addpath(paths["tools"], nargout=0)
+	subjects = os.listdir(paths["hrir"])
+	speakers = os.listdir(paths["timit"])
+
+	NUM_RECORDINGS = 100000
+
+	# start timer
+	start_time = time.time()
+
+	for i in range(NUM_RECORDINGS):
+		subject = random.choice(subjects)
+		speaker = random.choice(speakers)
+		elev_az = random.choice(combos)
+		filename = (subject.split("_")[1] + "_" + speaker[:-4] + 
+					"_" + str(elev_az[0]) + "_" + str(elev_az[1]) + ".wav")
+		eng.greg_synthesize(
+					(paths["hrir"] + subject),
+					(paths["timit"] + speaker),
+					elev_az[0],
+					elev_az[1],
+					(paths["save"] + filename),
+					nargout=0)
+
+	# stop timer
+	elapsed = time.time() - start_time
+	m, s = divmod(elapsed, 60)
+	h, m = divmod(m, 60)
+	print "Time elapsed: ", elapsed
+	print "Average time: ", elapsed / float(NUM_RECORDINGS)
+	print "Formatted: %d:%02d:%02d" % (h, m, s)
+
+main()

Data Processing/preprocessing.py

@@ -0,0 +1,52 @@
+"""
+Pre-processing script to transform WAV into STFT 
+for input into the CNN pipeline.
+"""
+
+from scipy import signal
+from scipy.io import wavfile
+import numpy
+import time
+import os
+
+def preprocess(filepath):
+	# CONSTANTS
+	WINDOW_LENGTH = 0.025
+	HOP_SIZE = 0.015
+	# read file, split into L-R, downsample, split into chunks 20 sample chunks
+	sample_rate, samples = wavfile.read(filepath)
+	samples_L = samples[:, 0]
+	samples_R = samples[:, 1]
+	# compute mag and phase spectrograms
+	nperseg = int(WINDOW_LENGTH / (1.0 / sample_rate))
+	noverlap = int(HOP_SIZE / (1.0 / sample_rate))
+	frequencies_L, times_L, spectogram_L = signal.spectrogram(samples_L, sample_rate, nperseg=nperseg, noverlap=noverlap, window="hamming", mode="complex")
+	frequencies_R, times_R, spectogram_R = signal.spectrogram(samples_R, sample_rate, nperseg=nperseg, noverlap=noverlap, window="hamming", mode="complex")
+	mag_spectrogram_L = numpy.absolute(spectogram_L)
+	mag_spectrogram_R = numpy.absolute(spectogram_R)
+	phase_spectrogram_L = numpy.angle(spectogram_L)
+	phase_spectrogram_R = numpy.angle(spectogram_R)
+	# return tensor 
+	spectrograms_combined = numpy.concatenate((mag_spectrogram_L, phase_spectrogram_L, mag_spectrogram_R, phase_spectrogram_R), axis=0)
+	spectrograms_combined = spectrograms_combined.astype("float16")
+	return spectrograms_combined
+
+recording_root = "/Users/ghunk/Desktop/GRADUATE/CSC_464/Final_Project/Dataset/binaural_random/"
+save_root = "/Users/ghunk/Desktop/GRADUATE/CSC_464/Final_Project/Dataset/stft/"
+
+def main():
+	start_time = time.time()
+	for i, recording in enumerate(os.listdir(recording_root)):
+		stft = preprocess(recording_root + recording)
+		filename = save_root + recording.split('.')[0]
+		numpy.save(filename, stft)
+		if ((i+1) % 100 == 0):
+			current_time = time.time() - start_time
+			print "Time elapsed:", current_time
+			print "Time per record:", current_time / (i+1)
+			print i+1, "records saved."
+	print "Finished."
+	print "Time elapsed:", time.time() - start_time
+
+main()
+

LICENSE.txt

@@ -0,0 +1,7 @@
+Copyright 2017 Gregory D. Hunkins
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

datagenerator.py Neural Net/datagenerator.py

@@ -2,14 +2,14 @@
 Main Neural Network Pipeline.
 """
 
-import numpy as np
 
 from keras.models import Sequential
 from keras.layers import Dense, Dropout, Flatten
 from keras.layers import Conv2D, Conv1D, GlobalAveragePooling2D
 from datagenerator import DataGenerator
 from sklearn.model_selection import train_test_split
 from sklearn.preprocessing import LabelEncoder
+import numpy as np
 import itertools
 import os
 
@@ -44,19 +44,12 @@
 # Design model
 model = Sequential()
 model.add(Conv2D(32, kernel_size=(3, 3), activation='relu', input_shape=(None, None, 1)))
-#model.add(Conv1D(256, kernel_size=(3), activation='relu', input_shape=(None, 1)))
 model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
 model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
 model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
-#model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
-#model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
-#model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
 model.add(GlobalAveragePooling2D(data_format='channels_last'))
-#model.add(Dropout(0.25))
 
-#model.add(Flatten())
 model.add(Dense(168, activation='relu'))
-#model.add(Dropout(0.25))
 model.add(Dense(168, activation='softmax'))
 
 model.compile(loss='categorical_crossentropy',
@@ -74,7 +67,3 @@
 
 #model.save_weights("/Users/ghunk/Desktop/GRADUATE/CSC_464/Final_Project/weights.h5py")
 model.save_weights("/home/ghunkins/weights.h5py")
-
-
-#steps_per_epoch = len(Train_IDs)//params['batch_size'],
-#validation_steps = len(Test_IDs)//params['batch_size'],

neuralnet.py Neural Net/neuralnet.py

@@ -1 +1,21 @@
-# Binaural-Source-Localization-CNN
+# Binaural-Source-Localization-CNN
+
+### Basic Information
+---
+**Author:** Gregory Hunkins
+
+**Organization:** University of Rochester
+
+**License:** MIT
+
+**Abstract:** A Convolutional Neural Network (CNN) classification system was designed for the task of source localization of human voices in 3-D space. A new dataset, VoiceBin100K, is introduced to accomplish this task and for future work in the field. The CNN inputs variable-length binaurual short- time Fourier Transform (STFT) magnitude and phase features and predicts location of the speaker’s voice according to 168 location classes. 
+
+
+### Running The Code
+---
+
+Reference: https://cs.rochester.edu/~cxu22/t/577F17/bluehive_tutorial.html
+
+### Data
+---
+Please contact [email protected] for access to the data. A public link will available shortly.