hasslesstech/neuro-lab8
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

incremental backup

Browse Source
This commit is contained in:
ІО-23 Шмуляр Олег 2025-12-09 08:09:49 +02:00
parent d998a04d47
commit 03712e07d8
4 changed files with 21 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
loss.py
View File

@ -9,5 +9,13 @@ def CTCLoss(y_true, y_pred):
input_length = input_length * tf.ones(shape = (batch_len, 1), dtype = "int64") input_length = input_length * tf.ones(shape = (batch_len, 1), dtype = "int64")
label_length = label_length * tf.ones(shape = (batch_len, 1), dtype = "int64") label_length = label_length * tf.ones(shape = (batch_len, 1), dtype = "int64")
#y_pred = tf.math.log_softmax(y_pred)
#y_true = tf.cast(y_true, dtype = "int64")
#y_true_sparse = tf.keras.backend.ctc_label_dense_to_sparse(y_true, label_length)
#print(y_true_sparse)
loss = keras.backend.ctc_batch_cost(y_true, y_pred, input_length, label_length) loss = keras.backend.ctc_batch_cost(y_true, y_pred, input_length, label_length)
#loss = tf.nn.ctc_loss(y_true_sparse, y_pred, input_length, label_length)
return loss return loss

6
model.py
View File

@ -1,3 +1,5 @@
import tensorflow as tf
from tensorflow.keras import layers as kl from tensorflow.keras import layers as kl
from tensorflow.keras import models as km from tensorflow.keras import models as km
from tensorflow.keras import losses as ks from tensorflow.keras import losses as ks
@ -8,6 +10,10 @@ from tensorflow import keras
from loss import CTCLoss from loss import CTCLoss
#g = tf.config.experimental.list_physical_devices('GPU')
#for i in g:
# tf.config.experimental.set_memory_growth(i, True)
def model(input_dim, output_dim, rnn_layers = 3, rnn_units = 72): def model(input_dim, output_dim, rnn_layers = 3, rnn_units = 72):
li = kl.Input((None, input_dim)) li = kl.Input((None, input_dim))

2
preprocessing.py
View File

@ -38,7 +38,7 @@ def encode_single_sample(wav, label):
def to_dataset(df, batch_size = 32): def to_dataset(df, batch_size = 32):
ds = tf.data.Dataset.from_tensor_slices((list(df["file_name"]), ds = tf.data.Dataset.from_tensor_slices((list(df["file_name"]),
list(df["normalized_transcription"]))) list(df["normalized_transcription"])))
ds = ds.map(encode_single_sample, num_parallel_calls=tf.data.AUTOTUNE) \ ds = ds.map(encode_single_sample, num_parallel_calls = tf.data.AUTOTUNE) \
.padded_batch(batch_size) \ .padded_batch(batch_size) \
.prefetch(buffer_size=tf.data.AUTOTUNE) .prefetch(buffer_size=tf.data.AUTOTUNE)

12
train.py
View File

@ -12,9 +12,9 @@ data = pd.read_csv("/mnt/tmpfs1/LJSpeech-1.1/metadata.csv",
header = None, header = None,
quoting = 3, quoting = 3,
names = ['file_name', 'i', 'normalized_transcription']) names = ['file_name', 'i', 'normalized_transcription'])
s = int(len(data) // 2 * 0.90) s = int(len(data) * 0.90)
train_data = data[:s] train_data = data[:s]
valid_data = data[s:len(data) // 2] valid_data = data[s:]
train_ds = to_dataset(train_data, batch_size = bs) train_ds = to_dataset(train_data, batch_size = bs)
valid_ds = to_dataset(valid_data, batch_size = bs) valid_ds = to_dataset(valid_data, batch_size = bs)
@ -22,13 +22,13 @@ valid_ds = to_dataset(valid_data, batch_size = bs)
m = model(input_dim = fft_length // 2 + 1, m = model(input_dim = fft_length // 2 + 1,
output_dim = char_to_num.vocabulary_size()) output_dim = char_to_num.vocabulary_size())
m.load_weights('model20-latest.keras') m.load_weights('model23-latest.keras')
ckpt1 = kc.ModelCheckpoint('model21-latest.keras', ckpt1 = kc.ModelCheckpoint('model24-latest.keras',
monitor = 'val_loss', monitor = 'val_loss',
save_best_only = False, save_best_only = False,
verbose = 1) verbose = 1)
ckpt2 = kc.ModelCheckpoint('model21-best.keras', ckpt2 = kc.ModelCheckpoint('model24-best.keras',
monitor = 'val_loss', monitor = 'val_loss',
save_best_only = True, save_best_only = True,
verbose = 1) verbose = 1)
@ -36,6 +36,6 @@ ckpt2 = kc.ModelCheckpoint('model21-best.keras',
ce1 = ce(valid_ds, m) ce1 = ce(valid_ds, m)
m.fit(train_ds, m.fit(train_ds,
epochs = 8, epochs = 80,
validation_data = valid_ds, validation_data = valid_ds,
callbacks = [ckpt1, ckpt2, ce1]) callbacks = [ckpt1, ckpt2, ce1])