File size: 1,745 Bytes
a445ec4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
import torch
from utils.dataset import Speech2Text, speech_collate_fn
from models.model import TransformerTransducer
# ==== Load Dataset ====
train_dataset = Speech2Text(
json_path="/home/anhkhoa/transformer_transducer/data/train.json",
vocab_path="/home/anhkhoa/transformer_transducer/data/vocab.json"
)
train_loader = torch.utils.data.DataLoader(
train_dataset,
batch_size=2,
shuffle=True,
collate_fn = speech_collate_fn
)
# ==== Kiểm tra 1 batch ====
batch = next(iter(train_loader))
# print("✅ Batch loaded!")
# print("Fbank shape :", batch['fbank'].shape) # [B, T, 80]
# print("Fbank lengths :", batch['fbank_len']) # [B]
# print("Text shape :", batch['text'].shape) # [B, U]
# print("Text lengths :", batch['text_len']) # [B]
# ==== Load model (giả sử bạn có config) ====
model = TransformerTransducer(
in_features=80,
n_classes=len(train_dataset.vocab),
n_layers=4,
n_dec_layers=2,
d_model=256,
ff_size=1024,
h=4,
joint_size=512,
enc_left_size=2,
enc_right_size=2,
dec_left_size=1,
dec_right_size=1,
p_dropout=0.1
)
def calculate_mask(lengths, max_len):
"""Tạo mask cho các tensor có chiều dài khác nhau"""
mask = torch.arange(max_len, device=lengths.device)[None, :] < lengths[:, None]
return mask
with torch.no_grad():
output, fbank_len, text_len = model(
speech=batch["fbank"], # [B, T, 80]
speech_mask=batch["fbank_mask"], # [B, T]
text=batch["text"], # [B, U]
text_mask=batch["text_mask"] # [B, U]
)
print("✅ Model output shape:", output.shape) # [B, T, U, vocab_size]
|