#! /usr/bin/python
#*******************************
import sys
import os
from os.path import join, isdir
from random import shuffle
import glob
from Load_sp_model import Load_sp_models
from Make_ASR_scp_text_format_fast import format_tokenize_data
import Attention_arg
from Attention_arg import parser
args = parser.parse_args()
if not isdir(args.data_dir):
os.makedirs(args.data_dir)
Word_model = Load_sp_models(args.Word_model_path)
Char_model = Load_sp_models(args.Char_model_path)
format_tokenize_data(scp_file=glob.glob(args.train_path + "*"),
transcript=args.text_file,
Translation=args.text_file,
outfile=open(join(args.data_dir, 'train_scp'), 'w'),
Word_model=Word_model,
Char_model=Char_model)
format_tokenize_data(scp_file=glob.glob(args.dev_path + "*"),
transcript=args.text_file,
Translation=args.text_file,
import json
from argparse import Namespace
#**********
from Initializing_model_LSTM_SS_v2_args import Initialize_Att_model
from Load_sp_model import Load_sp_models
from CMVN import CMVN
from utils__ import plotting
from user_defined_losses import compute_cer
from Decoding_loop import get_cer_for_beam
#-----------------------------------
import torch
import Attention_arg
from Attention_arg import parser
args = parser.parse_args()
model_path_name = join(args.model_dir, 'model_architecture_')
print(model_path_name)
#--------------------------------
###load the architecture if you have to load
with open(model_path_name, 'r') as f:
TEMP_args = json.load(f)
ns = Namespace(**TEMP_args)
args = parser.parse_args(namespace=ns)
if args.Am_weight < 1:
##model class
from RNNLM import RNNLM
##config file for RNLM