def __init__(self, path = None,serviceAddress='http://172.16.100.29:5000/recognize/',fun_use=False):
self.httpService = serviceAddress
#音频波形动态显示，实时显示波形，实时进行离散傅里叶变换分析频域
if path is not None and os.path.isfile(path):
self.stream = wave.open(path)
self.rate = self.stream.getparams()[2]
self.chunk = int(self.rate/1000*125)
self.read = self.stream.readframes
else:
self.rate = 16000
self.chunk = 2000
p = pyaudio.PyAudio()
self.stream = p.open(format=pyaudio.paInt16, channels=1, rate=self.rate,
input=True, frames_per_buffer=self.chunk)
self.read = self.stream.read
self.yysb = GatedConv.load("语音识别MASR/pretrained/gated-conv.pth")
self.data = []
fig = plt.figure(num='Real-time wave')
ax1 = fig.add_subplot(2, 1, 1)#两行一列，第一子图
ax2 = fig.add_subplot(2, 1, 2)#两行一列，第二子图
self.t = np.linspace(0, self.chunk - 1, self.chunk)
self.line1, = ax1.plot([], [], lw=2)
ax1.set_xlim(0, self.chunk)
ax1.set_ylim(-6000, 6000)
self.line2, = ax2.plot([], [], lw=2)
ax2.set_xlim(0, self.chunk)
ax2.set_ylim(-5000, 5000)
interval = int(1000*self.chunk/self.rate)#更新间隔/ms
if not fun_use:
animation.TimedAnimation.__init__(self, fig, interval=interval, blit=True)
def model_setup(args=None):
test_dataset = data.MASRDataset(args.test_index_path,
args.labels_path,
args.mode,
config=args)
dataloader = data.MASRDataLoader(test_dataset,
batch_size=args.batch_size,
num_workers=args.num_workers)
model = GatedConv.load(args.pretrained_path)
global decoder
decoder = BeamCTCDecoder(
dataloader.dataset.labels_str,
alpha=0.8,
beta=0.3,
lm_path="/root/lm/zh_giga.no_cna_cmn.prune01244.klm",
cutoff_top_n=40,
cutoff_prob=1.0,
beam_width=100,
num_processes=args.num_workers,
blank_index=0,
)
return model, dataloader
def model_setup(pretrained_path="pretrained/gated-conv.pth",
alpha=0.8,
beta=0.3,
lm_path="/kaggle/working/lm/zh_giga.no_cna_cmn.prune01244.klm",
cutoff_top_n=40,
cutoff_prob=1.0,
beam_width=32,
num_processes=4,
blank_index=0,
vocabulary=None):
num_processes = cpu_num
model = GatedConv.load(pretrained_path)
model.eval()
if vocabulary is not None:
model.vocabulary = vocabulary
decoder = CTCBeamDecoder(
model.vocabulary,
lm_path,
alpha,
beta,
cutoff_top_n,
cutoff_prob,
beam_width,
num_processes,
blank_index,
)
return model, decoder
def recognize():
datas = request.json
token = datas['token']
receipt_data = list(datas['data'])
if token == 'SR':
model = GatedConv.load("语音识别MASR/pretrained/gated-conv.pth")
text = model.predict(receipt_data)
return text
elif token == 'FN':
nums = np.array(receipt_data)
mean = np.mean(nums)
median = np.median(nums)
return '平均数：{} 中位数：{}'.format(mean, median)
def recognize():
datas = request.json
#datas = json.loads(request.get_data().decode("utf-8"))
token = datas['token']
receipt_data = list(datas['data'])
if token == 'SR':
model = GatedConv.load("AboutDL/语音识别MASR/pretrained/gated-conv.pth")
text = model.predict(receipt_data)
print(text)
return text
elif token == 'FN':
nums = np.array(receipt_data)
mean = np.mean(nums)
median = np.median(nums)
return '平均数：{} 中位数：{}'.format(mean,median)
import torch
import feature
from models.conv import GatedConv
import torch.nn.functional as F
from ctcdecode import CTCBeamDecoder
alpha = 0.8
beta = 0.3
lm_path = "lm/zh_giga.no_cna_cmn.prune01244.klm"
cutoff_top_n = 40
cutoff_prob = 1.0
beam_width = 32
num_processes = 4
blank_index = 0
model = GatedConv.load("pretrained/gated-conv.pth")
model.eval()
decoder = CTCBeamDecoder(
model.vocabulary,
lm_path,
alpha,
beta,
cutoff_top_n,
cutoff_prob,
beam_width,
num_processes,
blank_index,
)
import torch
import feature
from models.conv import GatedConv
import torch.nn.functional as F
from ctcdecode import CTCBeamDecoder
from config import lm_path, pretrained_model_path
alpha = 0.8
beta = 0.3
cutoff_top_n = 40
cutoff_prob = 1.0
beam_width = 32
num_processes = 4
blank_index = 0
model = GatedConv.load(pretrained_model_path)
model.eval()
decoder = CTCBeamDecoder(
model.vocabulary,
lm_path,
alpha,
beta,
cutoff_top_n,
cutoff_prob,
beam_width,
num_processes,
blank_index,
)
import _init_path
import platform
from models.conv import GatedConv
use_lm = True
if use_lm:
import beamdecode
system_type = platform.system()
if (system_type == 'Windows'):
model = GatedConv.load("AboutDL\\语音识别MASR\\pretrained\\gated-conv.pth")
#import scipy
#_,receipt_data = scipy.io.wavfile.read("E:\\打开欢呼比.wav")
#text = model.predict(receipt_data)事实证明效果相同
text = model.predict("E:\\打开欢呼比.wav")
elif (system_type == 'Linux'):
model = GatedConv.load('AboutDL/语音识别MASR/pretrained/gated-conv.pth')
text = model.predict(
"/media/yangjinming/DATA/Dataset/PrimeWords/d/d2/d25104a2-6be0-4950-9ec0-42e8e1303492.wav"
)
print("识别结果:", text)
import _init_path
from models.conv import GatedConv
model = GatedConv.load("语音识别MASR/pretrained/gated-conv.pth")
model.to_train()
model.fit("data/train.index", "data/dev.index", train_batch_size=2)
# import _init_path
from models.conv import GatedConv
# model = GatedConv.load("pretrained/gated-conv.pth")
# model = GatedConv.load("pretrained/model_62.pth")
model = GatedConv.load("pretrained2/model_81.pth")
text = model.predict("./sample_audio/8_16.wav")
print("")
print("识别结果:")
print(text)
__mtime__ = '20210318'
import os
from models.conv import GatedConv
from config import pretrained_model_path
model = GatedConv.load(os.path.join('..', pretrained_model_path))
text = model.predict("../data_aishell/BAC009S0765W0130.wav")
print("")
print("识别结果:")
print(text)
import torch
import feature
from models.conv import GatedConv
import torch.nn.functional as F
from ctcdecode import CTCBeamDecoder
alpha = 0.8
beta = 0.3
lm_path = "/home/db/bing/yuyingshibie/masr/lm/zh_giga.no_cna_cmn.prune01244.klm"
cutoff_top_n = 40
cutoff_prob = 1.0
beam_width = 32
num_processes = 4
blank_index = 0
model = GatedConv.load(
"/home/db/bing/yuyingshibie/masr/pretrained/gated-conv.pth")
model.eval()
decoder = CTCBeamDecoder(
model.vocabulary,
lm_path,
alpha,
beta,
cutoff_top_n,
cutoff_prob,
beam_width,
num_processes,
blank_index,
)
import _init_path
from models.conv import GatedConv
# model = GatedConv.load("pretrained/gated-conv.pth")
model = GatedConv.load("pretrained/model_3.pth")
text = model.predict("./sample_audio/test.wav")
print("")
print("识别结果:")
print(text)
parser.add_argument('--lm-alpha-from', default=0.0, type=float, help='Language model weight start tuning')
parser.add_argument('--lm-alpha-to', default=3.0, type=float, help='Language model weight end tuning')
parser.add_argument('--lm-beta-from', default=0.0, type=float,
help='Language model word bonus (all words) start tuning')
parser.add_argument('--lm-beta-to', default=0.5, type=float,
help='Language model word bonus (all words) end tuning')
parser.add_argument('--lm-num-alphas', default=45, type=float, help='Number of alpha candidates for tuning')
parser.add_argument('--lm-num-betas', default=5, type=float, help='Number of beta candidates for tuning')
parser = add_decoder_args(parser)
args = parser.parse_args()
if args.lm_path is None:
print("error: LM must be provided for tuning")
sys.exit(1)
model = GatedConv.load(args.model_path)
saved_output = np.load(args.saved_output, allow_pickle=True)
def init(beam_width, blank_index, lm_path):
global decoder, ae_decoder
decoder = BeamCTCDecoder(model.vocabulary, lm_path=lm_path, beam_width=beam_width, num_processes=args.lm_workers,
blank_index=blank_index)
ae_decoder = GreedyDecoder(model.vocabulary)
def decode_dataset(params):
lm_alpha, lm_beta = params
global decoder
decoder._decoder.reset_params(lm_alpha, lm_beta)