fixing sampling of start for clipping from noise audio

avsr/awgn.py

@@ -1,83 +1,83 @@
-import numpy as np
-from os import path
-from .audio import read_wav_file
-current_path = path.abspath(path.dirname(__file__))
-
-
-def generate_random_vector(length):
-    random_sig = np.random.normal(0, 1, length)
-    return np.expand_dims(random_sig, -1)
-
-
-def add_noise(orig_signal, noise_type, snr, sampling_rate):
-
-    signal_length = len(orig_signal)
-    sig_power = np.sum(np.abs(orig_signal)**2)/signal_length
-    target_snr = 10**(snr/10)
-
-    if noise_type == 'white':
-        noise_data = generate_random_vector(signal_length)
-    elif noise_type == 'cafe':
-        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'cafeteria_babble.wav'), sampling_rate)
-        noise_data = random_segment(noise_data, signal_length)
-        sig_power /= np.sum(np.abs(noise_data)**2)/signal_length
-    elif noise_type == 'street':
-        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'street_noise_downtown.wav'), sampling_rate)
-        noise_data = random_segment(noise_data, signal_length)
-        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
-    else:
-        raise Exception('unknown noise type, did you mean `white` ?')
-
-    noise_power = float(sig_power/target_snr)
-    noise_variance = np.sqrt(noise_power)
-
-    noise = noise_variance * noise_data
-
-    return np.expand_dims(orig_signal, axis=-1) + noise
-
-
-def add_noise_cached(orig_signal, noise_type, noise_data, snr):
-    signal_length = len(orig_signal)
-    sig_power = np.sum(np.abs(orig_signal) ** 2) / signal_length
-    target_snr = 10 ** (snr / 10)
-
-    if noise_type == 'wgn':
-        noise_data = generate_random_vector(signal_length)
-    elif noise_type == 'cafe':
-        noise_data = random_segment(noise_data, signal_length)
-        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
-    elif noise_type == 'street':
-        noise_data = random_segment(noise_data, signal_length)
-        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
-    else:
-        raise Exception('unknown noise type, did you mean `wgn` ?')
-
-    noise_power = float(sig_power / target_snr)
-    noise_variance = np.sqrt(noise_power)
-
-    noise = noise_variance * noise_data
-
-    return orig_signal + noise
-
-
-def random_segment(data, target_len):
-    from random import randint
-
-    start_limit = len(data) - target_len + 1
-    start = randint(0, start_limit)
-
-    return data[start:start + target_len]
-
-
-def cache_noise(noise_type, sampling_rate):
-
-    if noise_type == 'wgn':
-        noise_data = None
-    elif noise_type == 'cafe':
-        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'cafeteria_babble.wav'), sampling_rate)
-    elif noise_type == 'street':
-        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'street_noise_downtown.wav'), sampling_rate)
-    else:
-        raise Exception('unknown noise type, did you mean `white` ?')
-
-    return noise_data
+import numpy as np
+from os import path
+from .audio import read_wav_file
+current_path = path.abspath(path.dirname(__file__))
+
+
+def generate_random_vector(length):
+    random_sig = np.random.normal(0, 1, length)
+    return np.expand_dims(random_sig, -1)
+
+
+def add_noise(orig_signal, noise_type, snr, sampling_rate):
+
+    signal_length = len(orig_signal)
+    sig_power = np.sum(np.abs(orig_signal)**2)/signal_length
+    target_snr = 10**(snr/10)
+
+    if noise_type == 'white':
+        noise_data = generate_random_vector(signal_length)
+    elif noise_type == 'cafe':
+        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'cafeteria_babble.wav'), sampling_rate)
+        noise_data = random_segment(noise_data, signal_length)
+        sig_power /= np.sum(np.abs(noise_data)**2)/signal_length
+    elif noise_type == 'street':
+        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'street_noise_downtown.wav'), sampling_rate)
+        noise_data = random_segment(noise_data, signal_length)
+        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
+    else:
+        raise Exception('unknown noise type, did you mean `white` ?')
+
+    noise_power = float(sig_power/target_snr)
+    noise_variance = np.sqrt(noise_power)
+
+    noise = noise_variance * noise_data
+
+    return np.expand_dims(orig_signal, axis=-1) + noise
+
+
+def add_noise_cached(orig_signal, noise_type, noise_data, snr):
+    signal_length = len(orig_signal)
+    sig_power = np.sum(np.abs(orig_signal) ** 2) / signal_length
+    target_snr = 10 ** (snr / 10)
+
+    if noise_type == 'wgn':
+        noise_data = generate_random_vector(signal_length)
+    elif noise_type == 'cafe':
+        noise_data = random_segment(noise_data, signal_length)
+        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
+    elif noise_type == 'street':
+        noise_data = random_segment(noise_data, signal_length)
+        sig_power /= np.sum(np.abs(noise_data) ** 2) / signal_length
+    else:
+        raise Exception('unknown noise type, did you mean `wgn` ?')
+
+    noise_power = float(sig_power / target_snr)
+    noise_variance = np.sqrt(noise_power)
+
+    noise = noise_variance * noise_data
+
+    return orig_signal + noise
+
+
+def random_segment(data, target_len):
+    from random import randint
+
+    start_limit = len(data) - target_len
+    start = randint(0, start_limit)
+
+    return data[start:start + target_len]
+
+
+def cache_noise(noise_type, sampling_rate):
+
+    if noise_type == 'wgn':
+        noise_data = None
+    elif noise_type == 'cafe':
+        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'cafeteria_babble.wav'), sampling_rate)
+    elif noise_type == 'street':
+        noise_data = read_wav_file(path.join(current_path, 'noise_data', 'street_noise_downtown.wav'), sampling_rate)
+    else:
+        raise Exception('unknown noise type, did you mean `white` ?')
+
+    return noise_data