2022-09-18 20:56:14 +08:00
|
|
|
|
# !/usr/bin/env python3
|
2021-11-15 19:58:50 +08:00
|
|
|
|
# -*- coding: utf-8 -*-
|
|
|
|
|
|
#
|
|
|
|
|
|
# Copyright 2016-2099 Ailemon.net
|
|
|
|
|
|
#
|
|
|
|
|
|
# This file is part of ASRT Speech Recognition Tool.
|
|
|
|
|
|
#
|
|
|
|
|
|
# ASRT is free software: you can redistribute it and/or modify
|
|
|
|
|
|
# it under the terms of the GNU General Public License as published by
|
|
|
|
|
|
# the Free Software Foundation, either version 3 of the License, or
|
|
|
|
|
|
# (at your option) any later version.
|
|
|
|
|
|
# ASRT is distributed in the hope that it will be useful,
|
|
|
|
|
|
# but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
|
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
|
|
# GNU General Public License for more details.
|
|
|
|
|
|
#
|
|
|
|
|
|
# You should have received a copy of the GNU General Public License
|
|
|
|
|
|
# along with ASRT. If not, see <https://www.gnu.org/licenses/>.
|
|
|
|
|
|
# ============================================================================
|
|
|
|
|
|
|
|
|
|
|
|
"""
|
|
|
|
|
|
@author: nl8590687
|
|
|
|
|
|
用于训练语音识别系统语音模型的程序
|
|
|
|
|
|
"""
|
2022-09-18 20:56:14 +08:00
|
|
|
|
|
2021-11-15 19:58:50 +08:00
|
|
|
|
import os
|
|
|
|
|
|
from tensorflow.keras.optimizers import Adam
|
|
|
|
|
|
|
|
|
|
|
|
from speech_model import ModelSpeech
|
2022-05-24 00:02:28 +08:00
|
|
|
|
from model_zoo.speech_model.keras_backend import SpeechModel251BN
|
2021-11-15 19:58:50 +08:00
|
|
|
|
from data_loader import DataLoader
|
2022-03-27 21:47:12 +08:00
|
|
|
|
from speech_features import SpecAugment
|
2021-11-15 19:58:50 +08:00
|
|
|
|
|
|
|
|
|
|
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
|
|
|
|
|
|
2021-11-21 20:20:26 +08:00
|
|
|
|
AUDIO_LENGTH = 1600
|
|
|
|
|
|
AUDIO_FEATURE_LENGTH = 200
|
|
|
|
|
|
CHANNELS = 1
|
2021-11-15 19:58:50 +08:00
|
|
|
|
# 默认输出的拼音的表示大小是1428，即1427个拼音+1个空白块
|
2021-11-21 20:20:26 +08:00
|
|
|
|
OUTPUT_SIZE = 1428
|
2022-03-27 21:47:12 +08:00
|
|
|
|
sm251bn = SpeechModel251BN(
|
2021-11-21 20:20:26 +08:00
|
|
|
|
input_shape=(AUDIO_LENGTH, AUDIO_FEATURE_LENGTH, CHANNELS),
|
|
|
|
|
|
output_size=OUTPUT_SIZE
|
2022-09-18 20:56:14 +08:00
|
|
|
|
)
|
2022-03-27 21:47:12 +08:00
|
|
|
|
feat = SpecAugment()
|
2021-11-15 19:58:50 +08:00
|
|
|
|
train_data = DataLoader('train')
|
2022-09-18 20:56:14 +08:00
|
|
|
|
opt = Adam(learning_rate=0.0001, beta_1=0.9, beta_2=0.999, decay=0.0, epsilon=10e-8)
|
2022-03-27 21:47:12 +08:00
|
|
|
|
ms = ModelSpeech(sm251bn, feat, max_label_length=64)
|
2021-11-15 19:58:50 +08:00
|
|
|
|
|
2022-09-18 20:56:14 +08:00
|
|
|
|
# ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
|
2021-11-21 20:20:26 +08:00
|
|
|
|
ms.train_model(optimizer=opt, data_loader=train_data,
|
2022-09-18 20:56:14 +08:00
|
|
|
|
epochs=50, save_step=1, batch_size=16, last_epoch=0)
|
2022-03-27 21:47:12 +08:00
|
|
|
|
ms.save_model('save_models/' + sm251bn.get_model_name())
|
