reader = ScpStreamReader(args.data_scp,
mean_sub=args.mean_sub,
downsample=args.downsample)
reader.initialize()
space, beam_size, max_len = args.space, args.beam_size, args.max_len
start, win, stable_time = args.start_block, args.incl_block, args.stable_time
head, padding = args.attn_head, args.attn_padding
since = time.time()
fctm = open(args.output, 'w')
total_latency = 0
count = 0
with torch.no_grad():
while True:
utt, mat = reader.read_next_utt()
if utt is None or utt == '': break
print(utt)
src_seq = torch.FloatTensor(mat).to(device)
time_len = src_seq.size(0)
stable_hypo = [1]
latency = 0.
for i in range(max(time_len - start - 1, 0) // win + 1):
end = min(time_len, win * (i + 1) + start)
src = src_seq[0:end, :]
enc_out, mask, hypo, score, sth = partial_search(
model, src, beam_size, max_len, stable_hypo)
print(sth)
#!/usr/bin/env python3
# encoding: utf-8
# Copyright 2019 Thai-Son Nguyen
# Licensed under the Apache License, Version 2.0 (the "License")
import os
import argparse
from pynn.io.kaldi_seq import ScpStreamReader
parser = argparse.ArgumentParser(description='pynn')
parser.add_argument('--data-scp', help='path to data scp', required=True)
parser.add_argument('--output',
help='output file',
type=str,
default='data.len')
if __name__ == '__main__':
args = parser.parse_args()
loader = ScpStreamReader(args.data_scp)
loader.initialize()
fout = open(args.output, 'w')
while True:
utt_id, utt_mat = loader.read_next_utt()
if utt_id is None or utt_id == '': break
fout.write(utt_id + u' ' + str(len(utt_mat)) + os.linesep)
fout.close()