Initialize_new_benchmark

MLAgentBench/benchmarks/audio_signal_enhancement/env/create_config_files.py

@@ -0,0 +1,100 @@
+import random, glob
+
+def one_corpus(min_snr, max_snr, val_size, noise_len, train_val_noise_len, clip_len, \
+    noise_dpath, trainval_speech_dpath, test_speech_dpath, train_fname, val_fname, \
+    test_fname, seed):
+
+    # Fix the seed for the random number generator for reproducibility
+    random.seed(seed)
+
+    # noise file names
+    noise_fnames = glob.glob(noise_dpath + '**/*.[wW][aA][vV]', recursive=True)
+    noise_fnames = random.sample(noise_fnames, len(noise_fnames))
+
+    # speech file names for the training and the validation sets
+    speech_fnames = glob.glob(trainval_speech_dpath + '**/*.[wW][aA][vV]', recursive=True)
+    speech_fnames = random.sample(speech_fnames, len(speech_fnames))
+
+    # validation set
+    with open(val_fname, 'w') as f:
+        for i in range(val_size):
+            f.write(speech_fnames[i] + ' ' \
+                + noise_fnames[random.randint(0, len(noise_fnames) - 1)] + ' '\
+                    + str(random.uniform(0., train_val_noise_len - clip_len)) + ' '\
+                        + str(random.uniform(min_snr, max_snr)) + '\n')
+
+    # training set
+    with open(train_fname, 'w') as f:
+        for i in range(val_size, len(speech_fnames)):
+            f.write(speech_fnames[i] + ' ' \
+                + noise_fnames[random.randint(0, len(noise_fnames) - 1)] + ' '\
+                    + str(random.uniform(0., train_val_noise_len - clip_len)) + ' '\
+                        + str(random.uniform(min_snr, max_snr)) + ' '\
+                            + noise_fnames[random.randint(0, len(noise_fnames) - 1)] + ' '\
+                                + str(random.uniform(0., train_val_noise_len - clip_len)) + '\n')
+
+    # speech file names for the test set
+    speech_fnames = glob.glob(test_speech_dpath + '**/*.[wW][aA][vV]', recursive=True)
+    speech_fnames = random.sample(speech_fnames, len(speech_fnames))
+
+    # test set
+    with open(test_fname, 'w') as f:
+        for i in range(len(speech_fnames)):
+            f.write(speech_fnames[i] + ' ' \
+                + noise_fnames[random.randint(0, len(noise_fnames) - 1)] + ' '\
+                    + str(random.uniform(train_val_noise_len, noise_len - clip_len)) + ' '\
+                        + str(random.uniform(min_snr, max_snr)) + '\n')
+
+
+def main():
+
+    """
+    Create configuration files. Two sets of configuration files are created, which
+    correspond to TIMIT and voicebank. For TIMIT, TIMIT_train_set.txt, TIMIT_val_set.txt, 
+    and TIMIT_test_set.txt are created for the training, the validation, and the test sets. 
+    For voicebank, voicebank_train_set.txt, voicebank_val_set.txt, and voicebank_test_set.txt 
+    are created.
+    """
+
+    # TIMIT
+    min_snr = -5.
+    max_snr = 10.
+    val_size = 402
+    noise_len = 300.
+    train_val_noise_len = 270.
+    clip_len = 3.125 + 1e-7
+    noise_dpath = 'DEMAND/'
+    trainval_speech_dpath = 'TIMIT/TRAIN/'
+    test_speech_dpath = 'TIMIT/TEST/'
+    train_fname = 'TIMIT_train_set.txt'
+    val_fname = 'TIMIT_val_set.txt'
+    test_fname = 'TIMIT_test_set.txt'
+    seed = 10
+
+    one_corpus(min_snr, max_snr, val_size, noise_len, train_val_noise_len, clip_len, \
+        noise_dpath, trainval_speech_dpath, test_speech_dpath, train_fname, val_fname, \
+        test_fname, seed)
+
+    # voicebank
+    min_snr = -5.
+    max_snr = 10.
+    val_size = 1157
+    noise_len = 300.
+    train_val_noise_len = 270.
+    clip_len = 3.125 + 1e-7
+    noise_dpath = 'DEMAND/'
+    trainval_speech_dpath = 'voicebank/clean_trainset_wav'
+    test_speech_dpath = 'voicebank/clean_testset_wav'
+    train_fname = 'voicebank_train_set.txt'
+    val_fname = 'voicebank_val_set.txt'
+    test_fname = 'voicebank_test_set.txt'
+    seed = 20
+
+    one_corpus(min_snr, max_snr, val_size, noise_len, train_val_noise_len, clip_len, \
+        noise_dpath, trainval_speech_dpath, test_speech_dpath, train_fname, val_fname, \
+        test_fname, seed)
+
+
+if __name__ == '__main__':
+
+    main()