import tensorflow as tf import numpy as np import pandas as pd import hyperpara as hp from model import MyModel from midi_utils import midi_to_notes, notes_to_midi def predict_next_note( notes: np.ndarray, keras_model: MyModel, temperature: float = 1.0): """Generates a note IDs using a trained sequence model.""" # Add batch dimension inputs = tf.expand_dims(notes, 0) predictions = model.predict(inputs) pitch_logits = predictions['pitch'] step = predictions['step'] duration = predictions['duration'] pitch_logits /= temperature pitch = tf.random.categorical(pitch_logits, num_samples=1) pitch = tf.squeeze(pitch, axis=-1) duration = tf.squeeze(duration, axis=-1) step = tf.squeeze(step, axis=-1) step = tf.maximum(0, step) duration = tf.maximum(0, duration) return int(pitch), float(step), float(duration) def generate(sample_midi_file: str, output_midi_file: str, model, num_predictions): raw_notes = midi_to_notes(sample_midi_file)

sample_notes = np.stack([raw_notes[key] for key in hp.key_order], axis=1) input_notes = ( sample_notes[:hp.seq_length] / np.array([hp.vocab_size, 1, 1])) generated_notes = [] prev_start = 0 for _ in range(num_predictions): pitch, step, duration = predict_next_note( input_notes, model, hp.temperature) start = prev_start + step end = start + duration input_note = (pitch, step, duration) generated_notes.append((*input_note, start, end)) input_notes = np.delete(input_notes, 0, axis=0) input_notes = np.append( input_notes, np.expand_dims(input_note, 0), axis=0) prev_start = start generated_notes = pd.Dataframe( generated_notes, columns=(*hp.key_order, 'start', 'end')) print(generated_notes.head(10)) out_pm = notes_to_midi( generated_notes, out_file=output_midi_file, instrument_name="Acoustic Grand Piano") if __name__ == '__main__': model = MyModel() model.load_weights("training_checkpoints\ckpt_50")