def main(_): # REVIEW josephz: This paradigm was copied from inference-hack.py # initialize_globals() sample_dir = "sample" # sample_names = ["new_test"] sample_names = ["rolling_in_the_deep"] post_processor = PostProcessor() post_processor.load_weights("weights.h5") # sample_names = ["perfect_features"] # sample_names = ["rolling_in_the_one_more_time"] for sample_name in sample_names: console.h1("Processing %s" % sample_name) console.time("total processing for " + sample_name) sample_path = sample_dir + "/" + sample_name style_path = sample_path + "/style.mp3" content_path = sample_path + "/content.mp3" stylized_img_path = sample_path + "/stylized.png" stylized_img_raw_path = sample_path + "/stylized_raw.png" stylized_audio_path = sample_path + "/stylized.mp3" stylized_audio_raw_path = sample_path + "/stylized_raw.mp3" # Read style audio to spectrograms. style_audio, style_sample_rate = conversion.file_to_audio(style_path) style_img, style_phase = conversion.audio_to_spectrogram( style_audio, fft_window_size=1536) # Read content audio to spectrograms. content_audio, content_sample_rate = conversion.file_to_audio( content_path) content_img, content_phase = conversion.audio_to_spectrogram( content_audio, fft_window_size=1536) stylized_img_raw, stylized_img = stylize(content_img, style_img, content_phase, style_phase, content_path, style_path, post_processor) # Save raw stylized spectrogram and audio. stylized_audio_raw = conversion.amplitude_to_audio( stylized_img_raw, fft_window_size=1536, phase_iterations=15, phase=content_phase) conversion.image_to_file(stylized_img_raw, stylized_img_raw_path) conversion.audio_to_file(stylized_audio_raw, stylized_audio_raw_path) # Save post-processed stylized spectrogram and audio. stylized_audio = conversion.amplitude_to_audio(stylized_img, fft_window_size=1536, phase_iterations=15, phase=content_phase) # np.save("stylized_img.npy", stylized_img) # np.save("content_phase.npy", content_phase) conversion.image_to_file(stylized_img, stylized_img_path) conversion.audio_to_file(stylized_audio, stylized_audio_path) console.timeEnd("total processing for " + sample_name) console.info("Finished processing %s; saved to %s" % (sample_name, stylized_audio_path))
#!/usr/bin/env python import conversion import console import numpy as np from post_processor import PostProcessor post_processor = PostProcessor() post_processor.load_weights("weights.h5") stylized = conversion.file_to_image("sample/rolling_in_the_deep/stylized.png") content_harmonics = conversion.file_to_image("sample/rolling_in_the_deep/content.mp3.harmonics.png") content_sibilants = conversion.file_to_image("sample/rolling_in_the_deep/content.mp3.harmonics.png") stylized = post_processor.predict_unstacked(amplitude=stylized, harmonics=content_harmonics, sibilants=content_sibilants) conversion.image_to_file(stylized, "/Users/ollin/Desktop/boop.png")