def recognize(args):
model, LFR_m, LFR_n = Transformer.load_model(args.model_path)
print(model)
model.eval()
model.cuda()
char_list, sos_id, eos_id = process_dict(args.dict)
assert model.decoder.sos_id == sos_id and model.decoder.eos_id == eos_id
# read json data
with open(args.recog_json, 'rb') as f:
js = json.load(f)['utts']
# import Language Model
lm_model = kenlm.Model(args.lm_path)
# decode each utterance
new_js = {}
with torch.no_grad():
for idx, name in enumerate(js.keys(), 1):
print('(%d/%d) decoding %s' % (idx, len(js.keys()), name),
flush=True)
input = kaldi_io.read_mat(js[name]['input'][0]['feat']) # TxD
input = build_LFR_features(input, LFR_m, LFR_n)
input = torch.from_numpy(input).float()
input_length = torch.tensor([input.size(0)], dtype=torch.int)
input = input.cuda()
input_length = input_length.cuda()
nbest_hyps = model.recognize(input, input_length, char_list,
lm_model, args)
new_js[name] = add_results_to_json(js[name], nbest_hyps, char_list)
with open(args.result_label, 'wb') as f:
f.write(
json.dumps({
'utts': new_js
}, indent=4, sort_keys=True).encode('utf_8'))
def recognize(args):
# model
char_list, sos_id, eos_id = process_dict(args.dict)
vocab_size = len(char_list)
encoder = Encoder(
args.d_input * args.LFR_m,
args.n_layers_enc,
args.n_head,
args.d_k,
args.d_v,
args.d_model,
args.d_inner,
dropout=args.dropout,
pe_maxlen=args.pe_maxlen,
)
decoder = Decoder(
sos_id,
eos_id,
vocab_size,
args.d_word_vec,
args.n_layers_dec,
args.n_head,
args.d_k,
args.d_v,
args.d_model,
args.d_inner,
dropout=args.dropout,
tgt_emb_prj_weight_sharing=args.tgt_emb_prj_weight_sharing,
pe_maxlen=args.pe_maxlen,
)
model = Transformer(encoder, decoder)
model.load_state_dict(flow.load(args.model_path))
device = flow.device("cuda")
model.eval()
model.to(device)
LFR_m = args.LFR_m
LFR_n = args.LFR_n
char_list, sos_id, eos_id = process_dict(args.dict)
assert model.decoder.sos_id == sos_id and model.decoder.eos_id == eos_id
# read json data
with open(args.recog_json, "rb") as f:
js = json.load(f)["utts"]
# decode each utterance
new_js = {}
with flow.no_grad():
for idx, name in enumerate(js.keys(), 1):
print("(%d/%d) decoding %s" % (idx, len(js.keys()), name), flush=True)
input = kaldi_io.read_mat(js[name]["input"][0]["feat"])
input = build_LFR_features(input, LFR_m, LFR_n)
input = flow.tensor(input).to(dtype=flow.float32)
input_length = flow.tensor([input.size(0)], dtype=flow.int64)
input = input.to(device)
input_length = input_length.to(device)
nbest_hyps = model.recognize(input, input_length, char_list, args)
new_js[name] = add_results_to_json(js[name], nbest_hyps, char_list)
with open(args.result_label, "wb") as f:
f.write(json.dumps({"utts": new_js}, indent=4, sort_keys=True).encode("utf_8"))
def recognize(args):
model, LFR_m, LFR_n = Transformer.load_model(args.model_path, args)
print(model)
# device = torch.device("cuda:0")
# model = model.to(device)
# model = torch.nn.DataParallel(model, device_ids=list(range(args.ngpu)))
model.eval()
model.cuda()
char_list, sos_id, eos_id = process_dict(args.dict)
assert model.decoder.sos_id == sos_id and model.decoder.eos_id == eos_id
# read json data
with open(args.recog_json, 'r', encoding='utf-8') as f:
js = json.load(f)['utts']
# decode each utterance
new_js = {}
with torch.no_grad():
for idx, name in enumerate(js.keys(), 1):
print('(%d/%d) decoding %s' % (idx, len(js.keys()), name),
flush=True)
input = kaldi_io.read_mat(js[name]['input'][0]['feat']) # TxD
input = build_LFR_features(input, LFR_m, LFR_n)
input = torch.from_numpy(input).float()
input_length = torch.tensor([input.size(0)], dtype=torch.int)
input = input.cuda()
input_length = input_length.cuda()
nbest_hyps = model.recognize(input, input_length, char_list, args)
new_js[name] = add_results_to_json(js[name], nbest_hyps, char_list)
# new_js[name]['output'][0]['rec_text'] = new_js[name]['output'][0]['rec_text'].encode('utf_8')
# new_js[name]['output'][0]['rec_token'] = new_js[name]['output'][0]['rec_token'].encode('utf_8')
# new_js[name]['output'][0]['text'] = new_js[name]['output'][0]['text'].encode('utf_8')
# new_js[name]['output'][0]['token'] = new_js[name]['output'][0]['token'].encode('utf_8')
# print(new_js[name])
with open(args.result_label, 'wb') as f:
f.write(
json.dumps({
'utts': new_js
},
indent=4,
sort_keys=True,
ensure_ascii=False).encode('utf_8'))
def recognize(args):
#import pdb
#pdb.set_trace()
char_list, sos_id, eos_id = process_dict(args.dict)
args.char_list = char_list
model, LFR_m, LFR_n = Seq2Seq.load_model(args.model_path, args)
print(model)
model.eval()
model.cuda()
char_list, sos_id, eos_id = process_dict(args.dict)
assert model.decoder.sos_id == sos_id and model.decoder.eos_id == eos_id
# read json data
with open(args.recog_json, 'rb') as f:
js = json.load(f)['utts']
# decode each utterance
new_js = {}
with torch.no_grad():
for idx, name in enumerate(js.keys(), 1):
print('(%d/%d) decoding %s' % (idx, len(js.keys()), name),
flush=True)
input = kaldi_io.read_mat(js[name]['input'][0]['feat']) # TxD
input = build_LFR_features(input, LFR_m, LFR_n)
input = torch.from_numpy(input).float()
input_length = torch.tensor([input.size(0)], dtype=torch.int)
input = input.cuda()
input_length = input_length.cuda()
if args.align_trun:
align = (js[name]['output'][0]['ctcid'].split())
nbest_hyps = model.recognize_align(input, input_length,
char_list, align, args)
else:
nbest_hyps = model.recognize(input, input_length, char_list,
args)
new_js[name] = add_results_to_json(js[name], nbest_hyps, char_list)
with open(args.result_label, 'wb') as f:
f.write(
json.dumps({
'utts': new_js
}, indent=4, sort_keys=True).encode('utf_8'))
