Fehler as_list() beim Versuch, mein Audioklassifizierungsmodell zu trainieren ⇐ Python

[Guest]

Ich habe diesen Google Colab-Code und möchte ein Modell für MP3-Dateien mit einer Länge von 3 Sekunden, 44,10 kHz, 47,8 KB und 128 KBit/s trainieren, die sich in 6 Ordnern befinden, und ich erhalte immer wieder diese Fehlermeldung, wenn ich versuche zu trainieren :

Code: Select all

        ValueError                                Traceback (most recent call last)
 in ()
----> 1 hist = model.fit(train, epochs=4, validation_data=test)

1 frames
/usr/local/lib/python3.10/dist-packages/keras/engine/training.py in tf__train_function(iterator)
13                 try:
14                     do_return = True
---> 15                     retval_ = ag__.converted_call(ag__.ld(step_function), (ag__.ld(self), ag__.ld(iterator)), None, fscope)
16                 except:
17                     do_return = False

ValueError: in user code:

File "/usr/local/lib/python3.10/dist-packages/keras/engine/training.py", line 1160, in train_function  *
return step_function(self, iterator)
File "/usr/local/lib/python3.10/dist-packages/keras/engine/training.py", line 1146, in step_function  **
outputs = model.distribute_strategy.run(run_step, args=(data,))
File "/usr/local/lib/python3.10/dist-packages/keras/engine/training.py", line 1135, in run_step  **
outputs = model.train_step(data)
File "/usr/local/lib/python3.10/dist-packages/keras/engine/training.py", line 998, in train_step
return self.compute_metrics(x, y, y_pred, sample_weight)
File "/usr/local/lib/python3.10/dist-packages/keras/engine/training.py", line 1092, in compute_metrics
self.compiled_metrics.update_state(y, y_pred, sample_weight)
File "/usr/local/lib/python3.10/dist-packages/keras/engine/compile_utils.py", line 577, in update_state
self.build(y_pred, y_true)
File "/usr/local/lib/python3.10/dist-packages/keras/engine/compile_utils.py", line 483, in build
self._metrics = tf.__internal__.nest.map_structure_up_to(
File "/usr/local/lib/python3.10/dist-packages/keras/engine/compile_utils.py", line 631, in _get_metric_objects
return [self._get_metric_object(m, y_t, y_p) for m in metrics]
File "/usr/local/lib/python3.10/dist-packages/keras/engine/compile_utils.py", line 631, in 
return [self._get_metric_object(m, y_t, y_p) for m in metrics]
File "/usr/local/lib/python3.10/dist-packages/keras/engine/compile_utils.py", line 652, in _get_metric_object
y_t_rank = len(y_t.shape.as_list())

ValueError: as_list() is not defined on an unknown TensorShape.

DAS IST MEIN CODE und ich weiß nicht, was falsch ist. Ich hatte einen Fehler beim Laden des Sounds, aber jetzt funktioniert es, weil ich ein tf.Tensor-Objekt anstelle eines Strings erhalten habe. Jetzt weiß ich nicht warum Es funktioniert nicht. Ich verwende Tensoflow 2.10.0 und librosa:

Code: Select all

from google.colab import drive
drive.mount('/content/drive')

!pip install librosa tensorflow==2.10.0

import tensorflow as tf
import librosa
import os
import joblib
import numpy as np
from matplotlib import pyplot as plt

# Parametri globali
BATCH_SIZE = 4
SAMPLE_RATE = 44100  # Fișierele tale sunt 44.1 kHz
TARGET_SAMPLE_RATE = 16000  # Resamplează la 16 kHz
DURATION = 3  # Durata fișierului audio în secunde
N_MELS = 300  # Număr de benzi Mel
N_FFT = 1024  # Număr de puncte FFT
HOP_LENGTH = int((TARGET_SAMPLE_RATE * DURATION) / 1400)  # Calculăm hop_length pentru a obține 1400 frame-uri

# Funcție pentru încărcarea și preprocesarea fișierelor MP3
def load_sound(filename):

file_path = filename
if (isinstance(filename,tf.Tensor)):
file_path = filename.numpy().decode('utf-8')

# Încărcăm fișierul audio MP3 cu Librosa
print("F name: ", file_path)  # Verifică că avem path-ul corect
wav, sr = librosa.load(file_path, sr=TARGET_SAMPLE_RATE)

# Padding la o durată fixă de 3 secunde
wav = librosa.util.fix_length(wav, size=TARGET_SAMPLE_RATE * DURATION)

return wav

# Funcție pentru crearea spectrogramelor Mel
def create_spectrogram(file_path, label):
# Încărcăm fișierul audio
wav = load_sound(file_path)

# Calculăm spectrograma Mel
mel_spectrogram = librosa.feature.melspectrogram(
y=wav, sr=TARGET_SAMPLE_RATE, n_mels=N_MELS, n_fft=N_FFT, hop_length=HOP_LENGTH, fmax=8000
)

# Convertim la scară logaritmică (dB)
mel_spectrogram = librosa.power_to_db(mel_spectrogram, ref=np.max)

# Adăugăm o dimensiune suplimentară pentru intrarea modelului
mel_spectrogram = np.expand_dims(mel_spectrogram, axis=-1)

return mel_spectrogram, label

# Directorul de date
base_dir = '/content/drive/MyDrive/house_sounds/sound_data_1200'
folders = ['door', 'voice', 'glass', 'silence', 'dog', 'footsteps']

# Creăm lista de fișiere
files = []
for folder in folders:
folder_path = os.path.join(base_dir, folder)
file_paths = tf.data.Dataset.list_files(os.path.join(folder_path, '*.mp3'))
files.append(file_paths)

# Creăm dataset-ul
door = tf.data.Dataset.zip((files[0], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[0])], 0))))
voice = tf.data.Dataset.zip((files[1], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[1])], 1))))
glass = tf.data.Dataset.zip((files[2], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[2])], 2))))
silence = tf.data.Dataset.zip((files[3], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[3])], 3))))
dog = tf.data.Dataset.zip((files[4], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[4])], 4))))
footsteps = tf.data.Dataset.zip((files[5], tf.data.Dataset.from_tensor_slices(tf.fill([len(files[5])], 5))))

data = door.concatenate(voice).concatenate(glass).concatenate(silence).concatenate(dog).concatenate(footsteps)

total_files = sum([len(files) for files in files])
print(f"Total fișiere: {total_files}")

# Verificăm un exemplu
f_path, l_abel = data.shuffle(buffer_size=10000).as_numpy_iterator().next()
spectrogram, label = create_spectrogram(f_path, l_abel)
plt.figure(figsize=(30, 10))
plt.imshow(spectrogram[:, :, 0], cmap='viridis', aspect='auto')
plt.title(folders[label])
plt.show()

def process_data(file_path, label):
spectrogram, label = tf.py_function(func=create_spectrogram, inp=[file_path, label], Tout=(tf.float32, tf.int32))
spectrogram.set_shape([N_MELS, 1400, 1])  # Asigură-te că setăm dimensiunile

return spectrogram, label

data = data.map(process_data)
data = data.cache()
data = data.shuffle(buffer_size=5000)
data = data.batch(BATCH_SIZE)
data = data.prefetch(8)

print('Len:', len(data))
print('Cardinality:', data.cardinality().numpy())

train = data.take(2300)
test = data.skip(2300).take(150)

for spectrogram, label in train.skip(2299).take(1):   # Verifică un lot din train
print("Spectrogram shape:", spectrogram.shape)
print("Label shape:", label.shape)

# Modelul CNN
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, Dense, Flatten, MaxPooling2D

model = Sequential([
Conv2D(16, (3, 3), activation='relu', input_shape=(N_MELS, 1400, 1)),
MaxPooling2D(pool_size=(3, 3)),
Conv2D(32, (3, 3), activation='relu'),
MaxPooling2D(pool_size=(3, 3)),
Conv2D(64, (3, 3), activation='relu'),
MaxPooling2D(pool_size=(3, 3)),
Flatten(),
Dense(32, activation='relu'),
Dense(6, activation='softmax')
])

# Compilăm modelul
model.compile(
optimizer='Adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(),
metrics=['accuracy']
)

# Antrenăm modelul
model.summary()

hist = model.fit(train, epochs=4, validation_data=test)

# Salvăm modelul și etichetele
model.save('/content/drive/MyDrive/house_sounds/mel_librosa_1400x300_model')
model.save_weights('/content/drive/MyDrive/house_sounds/mel_librosa_1400x300_weights.h5')
joblib.dump(folders, '/content/drive/MyDrive/house_sounds/4_class_labels.pkl')

# Evaluăm modelul
test_loss, test_accuracy = model.evaluate(test)
print(f"Test Loss: {test_loss}, Test Accuracy: {test_accuracy}")