def output(file_count):
import level1
files = level1.get_files_data()
file = files[0]
file = file.reshape(len(file), level1.INPUT_COUNT, 1)
del files
model = level1.Level1Model()
print(file.shape)
file = model.encode(file)
print(file.shape)
del model
backend.clear_session()
model = Level2Model()
file = model.predict_output(file)
del model
backend.clear_session()
model = level1.Level1Model()
file = model.decode(file)
del model
backend.clear_session()
write_wav(get_output_path(f"output-level2-0.wav"), file.flatten())
def output(file_count):
files = get_files_data()
model = Level1Model()
for i in range(min(len(files), file_count)):
inp = files[i].reshape(len(files[i]), INPUT_COUNT, 1)
output = model.predict_output(inp).flatten()
write_wav(get_output_path(f"output-level1-{i}.wav"), output)
def visualize(model_cls, input_data):
os.environ["FFMPEG_BINARY"] = "ffmpeg"
model = model_cls()
output = model.encode(input_data)
output = output.reshape(output.shape[0] * 512, 128)
min_val = np.amin(output)
max_val_normalized = np.amax(output) - min_val
last_percentage = -1
figures = []
# (graph total duration / graph datapoint count) * (graph datapoint count / graph width)
figure_snapshot_rate = 40
tick_to_sample_ratio = 32.87890625 # This is still off sync with the audio, 2:53 becomes 2:58 for some reason
frame_duration = (figure_snapshot_rate * tick_to_sample_ratio) / 44100
for i in range(128):
column = i % 16
row = int(i / 16)
figures.append(Figure(60, 60, row, column, frame_duration))
print(f"Rendering output: {output.shape}")
for index, entry in enumerate(output):
should_snapshot = index % figure_snapshot_rate == 0
for plot_index, plot in enumerate(figures):
plot.push((entry[plot_index] - min_val) / max_val_normalized)
if should_snapshot:
plot.snapshot()
percentage = int(index / len(output) * 100)
if percentage % 1 == 0 and last_percentage != percentage:
last_percentage = percentage
print(f"Capturing figures: {percentage}%...")
print(f"{len(figures[0].figures)} figure frames rendered")
clips = [FigureClip(figure) for figure in figures]
audio_filename = f"vis/output.wav"
output = model.predict_output(input_data).flatten()
write_wav(audio_filename, output)
del model
backend.clear_session()
audio = AudioFileClip(audio_filename)
audio = audio.set_start(0)
audio = audio.set_duration(
min(audio.duration, frame_duration * len(figures[0].figures)))
result = CompositeVideoClip(clips, size=(16 * 66 + 12, 8 * 66 + 12))
result = result.set_audio(audio)
result.write_videofile("vis/output.mp4", fps=1 / frame_duration)
def output(file_count):
import level1
import level2
model1 = level1.Level1Model()
model2 = level2.Level2Model()
model3 = Level3Model()
files = level1.get_files_data()
file = files[0]
file = file.reshape(len(file), level1.INPUT_COUNT, 1)
del files
file = model1.encode(file)
file = model2.encode(file)
file = model3.predict_output(file)
file = model2.decode(file)
file = model1.decode(file)
write_wav(get_output_path(f"output-level3-0.wav"), file.flatten())