![]() ![]() Model.add(MaxPooling1D(pool_size=4, padding='same')) Model.add(Conv1D(30, kernel_size=4, activation='relu')) Model.add(Reshape((int(input_length / 13), 13), input_shape=(input_length, ))) Model.add(InputLayer(input_shape=(input_length, ), name='x_input')) Model.fit(train_dataset, epochs=300, validation_data=validation_dataset,class_weight=class_weights, verbose=2, callbacks=callbacks)įYI : the full definition of my NN model import tensorflow as tfįrom import Sequentialįrom import Dense, InputLayer, Dropout, Conv1D, Conv2D, Flatten, Reshape, MaxPooling1D, AveragePooling2D, BatchNormalizationįrom import Adamįrom _weight import compute_class_weight I appreciated the zoom functionality of each software but wanted auto-chunking or scrolling of the spectrogram within a high-resolution video.My data set is not very balanced so I have added the following to my NN architecture: from _weight import compute_class_weightĬlass_weights = dict(enumerate(compute_class_weight('balanced', np.unique(np.argmax(Y_train, axis=1)), np.argmax(Y_train, axis=1)))) Not having a seamless solution other than using screen capture on iZotope RX or Audacity spectrograms, I did some more research looking at iAnalyse 5 software (replaces eAnalysis software) and Cornell Lab’s RavenLite software, but was unsatisfied with movie export results. I’ve been frustrated with multiple software solutions to generating scrolling spectrogram videos. I was blown away by Merlin ID’s auto spectrogram video app, and I wanted to relook at how I create my own spectrogram videos. This led me to revisit my work with spectrograms. Yet, after an interview with a UO visiting researcher, Ines Moran, as part of our Soundscapes of Socioecological Succession project, I was further intrigued by how sound, spectrograms, and AI plays an integral role in her bioacoustics research on bird communication. I knew of the theories around animals occupying various frequency spaces within a habitat based upon the bioacoustics work of Garth Paine and great multimedia reporting by Andreas von Bubnoff. ![]() Yet, I am still amazed by the work of spectrograms and sound in the sciences. I’ve worked with time-frequency plots in various ways in the past, namely spectral smoothing music ( listen on Spotify), collaborative research ( read the paper), and even teaching ( Data Sonification course) at the University of Oregon. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |