用于测试语音识别系统语音模型的程序
"""
import os
from speech_model import ModelSpeech
from model_zoo.speech_model.keras_backend import SpeechModel251BN
from data_loader import DataLoader
from speech_features import Spectrogram
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
AUDIO_LENGTH = 1600
AUDIO_FEATURE_LENGTH = 200
CHANNELS = 1
# 默认输出的拼音的表示大小是1428，即1427个拼音+1个空白块
OUTPUT_SIZE = 1428
sm251bn = SpeechModel251BN(input_shape=(AUDIO_LENGTH, AUDIO_FEATURE_LENGTH,
CHANNELS),
output_size=OUTPUT_SIZE)
feat = Spectrogram()
evalue_data = DataLoader('dev')
ms = ModelSpeech(sm251bn, feat, max_label_length=64)
ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
ms.evaluate_model(data_loader=evalue_data,
data_count=-1,
out_report=True,
show_ratio=True,
show_per_step=100)
type=str,
help='the port to listen')
args = parser.parse_args()
app = Flask("ASRT API Service")
AUDIO_LENGTH = 1600
AUDIO_FEATURE_LENGTH = 200
CHANNELS = 1
# 默认输出的拼音的表示大小是1428，即1427个拼音+1个空白块
OUTPUT_SIZE = 1428
sm251bn = SpeechModel251BN(input_shape=(AUDIO_LENGTH, AUDIO_FEATURE_LENGTH,
CHANNELS),
output_size=OUTPUT_SIZE)
feat = Spectrogram()
ms = ModelSpeech(sm251bn, feat, max_label_length=64)
ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
ml = ModelLanguage('model_language')
ml.load_model()
class AsrtApiResponse:
'''
ASRT语音识别基于HTTP协议的API接口响应类
'''
def __init__(self, status_code, status_message='', result=''):
self.status_code = status_code
self.status_message = status_message
self.result = result
import os
from speech_model import ModelSpeech
from speech_model_zoo import SpeechModel251BN
from speech_features import Spectrogram
from language_model3 import ModelLanguage
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
AUDIO_LENGTH = 1600
AUDIO_FEATURE_LENGTH = 200
CHANNELS = 1
# 默认输出的拼音的表示大小是1428，即1427个拼音+1个空白块
OUTPUT_SIZE = 1428
sm251bn = SpeechModel251BN(input_shape=(AUDIO_LENGTH, AUDIO_FEATURE_LENGTH,
CHANNELS),
output_size=OUTPUT_SIZE)
feat = Spectrogram()
ms = ModelSpeech(sm251bn, feat, max_label_length=64)
ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
res = ms.recognize_speech_from_file('filename.wav')
print('*[提示] 声学模型语音识别结果：\n', res)
ml = ModelLanguage('model_language')
ml.load_model()
str_pinyin = res
res = ml.pinyin_to_text(str_pinyin)
print('语音识别最终结果：\n', res)
from tensorflow.keras.optimizers import Adam
from speech_model import ModelSpeech
from speech_model_zoo import SpeechModel251BN
from data_loader import DataLoader
from speech_features import SpecAugment
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
AUDIO_LENGTH = 1600
AUDIO_FEATURE_LENGTH = 200
CHANNELS = 1
# 默认输出的拼音的表示大小是1428，即1427个拼音+1个空白块
OUTPUT_SIZE = 1428
sm251bn = SpeechModel251BN(input_shape=(AUDIO_LENGTH, AUDIO_FEATURE_LENGTH,
CHANNELS),
output_size=OUTPUT_SIZE)
feat = SpecAugment()
train_data = DataLoader('train')
opt = Adam(lr=0.0001, beta_1=0.9, beta_2=0.999, decay=0.0, epsilon=10e-8)
ms = ModelSpeech(sm251bn, feat, max_label_length=64)
#ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
ms.train_model(optimizer=opt,
data_loader=train_data,
epochs=50,
save_step=1,
batch_size=16,
last_epoch=0)
ms.save_model('save_models/' + sm251bn.get_model_name())