|
|
@ -3,6 +3,7 @@ from __future__ import absolute_import
|
|
|
|
from __future__ import division
|
|
|
|
from __future__ import division
|
|
|
|
from __future__ import print_function
|
|
|
|
from __future__ import print_function
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
import sys
|
|
|
|
import numpy as np
|
|
|
|
import numpy as np
|
|
|
|
import argparse
|
|
|
|
import argparse
|
|
|
|
import functools
|
|
|
|
import functools
|
|
|
@ -10,32 +11,35 @@ import paddle.v2 as paddle
|
|
|
|
import _init_paths
|
|
|
|
import _init_paths
|
|
|
|
from data_utils.data import DataGenerator
|
|
|
|
from data_utils.data import DataGenerator
|
|
|
|
from model_utils.model import DeepSpeech2Model
|
|
|
|
from model_utils.model import DeepSpeech2Model
|
|
|
|
from utils.error_rate import wer
|
|
|
|
from utils.error_rate import wer, cer
|
|
|
|
from utils.utility import add_arguments, print_arguments
|
|
|
|
from utils.utility import add_arguments, print_arguments
|
|
|
|
|
|
|
|
|
|
|
|
parser = argparse.ArgumentParser(description=__doc__)
|
|
|
|
parser = argparse.ArgumentParser(description=__doc__)
|
|
|
|
add_arg = functools.partial(add_arguments, argparser=parser)
|
|
|
|
add_arg = functools.partial(add_arguments, argparser=parser)
|
|
|
|
# yapf: disable
|
|
|
|
# yapf: disable
|
|
|
|
add_arg('num_samples', int, 100, "# of samples to infer.")
|
|
|
|
add_arg('num_batches', int, -1, "# of batches tuning on. "
|
|
|
|
|
|
|
|
"Default -1, on whole dev set.")
|
|
|
|
|
|
|
|
add_arg('batch_size', int, 256, "# of samples per batch.")
|
|
|
|
add_arg('trainer_count', int, 8, "# of Trainers (CPUs or GPUs).")
|
|
|
|
add_arg('trainer_count', int, 8, "# of Trainers (CPUs or GPUs).")
|
|
|
|
add_arg('beam_size', int, 500, "Beam search width.")
|
|
|
|
add_arg('beam_size', int, 500, "Beam search width.")
|
|
|
|
add_arg('num_proc_bsearch', int, 12, "# of CPUs for beam search.")
|
|
|
|
add_arg('num_proc_bsearch', int, 12, "# of CPUs for beam search.")
|
|
|
|
add_arg('num_conv_layers', int, 2, "# of convolution layers.")
|
|
|
|
add_arg('num_conv_layers', int, 2, "# of convolution layers.")
|
|
|
|
add_arg('num_rnn_layers', int, 3, "# of recurrent layers.")
|
|
|
|
add_arg('num_rnn_layers', int, 3, "# of recurrent layers.")
|
|
|
|
add_arg('rnn_layer_size', int, 2048, "# of recurrent cells per layer.")
|
|
|
|
add_arg('rnn_layer_size', int, 2048, "# of recurrent cells per layer.")
|
|
|
|
add_arg('num_alphas', int, 14, "# of alpha candidates for tuning.")
|
|
|
|
add_arg('num_alphas', int, 45, "# of alpha candidates for tuning.")
|
|
|
|
add_arg('num_betas', int, 20, "# of beta candidates for tuning.")
|
|
|
|
add_arg('num_betas', int, 8, "# of beta candidates for tuning.")
|
|
|
|
add_arg('alpha_from', float, 0.1, "Where alpha starts tuning from.")
|
|
|
|
add_arg('alpha_from', float, 1.0, "Where alpha starts tuning from.")
|
|
|
|
add_arg('alpha_to', float, 0.36, "Where alpha ends tuning with.")
|
|
|
|
add_arg('alpha_to', float, 3.2, "Where alpha ends tuning with.")
|
|
|
|
add_arg('beta_from', float, 0.05, "Where beta starts tuning from.")
|
|
|
|
add_arg('beta_from', float, 0.1, "Where beta starts tuning from.")
|
|
|
|
add_arg('beta_to', float, 1.0, "Where beta ends tuning with.")
|
|
|
|
add_arg('beta_to', float, 0.45, "Where beta ends tuning with.")
|
|
|
|
add_arg('cutoff_prob', float, 0.99, "Cutoff probability for pruning.")
|
|
|
|
add_arg('cutoff_prob', float, 1.0, "Cutoff probability for pruning.")
|
|
|
|
|
|
|
|
add_arg('cutoff_top_n', int, 40, "Cutoff number for pruning.")
|
|
|
|
add_arg('use_gru', bool, False, "Use GRUs instead of simple RNNs.")
|
|
|
|
add_arg('use_gru', bool, False, "Use GRUs instead of simple RNNs.")
|
|
|
|
add_arg('use_gpu', bool, True, "Use GPU or not.")
|
|
|
|
add_arg('use_gpu', bool, True, "Use GPU or not.")
|
|
|
|
add_arg('share_rnn_weights',bool, True, "Share input-hidden weights across "
|
|
|
|
add_arg('share_rnn_weights',bool, True, "Share input-hidden weights across "
|
|
|
|
"bi-directional RNNs. Not for GRU.")
|
|
|
|
"bi-directional RNNs. Not for GRU.")
|
|
|
|
add_arg('tune_manifest', str,
|
|
|
|
add_arg('tune_manifest', str,
|
|
|
|
'data/librispeech/manifest.dev',
|
|
|
|
'data/librispeech/manifest.dev-clean',
|
|
|
|
"Filepath of manifest to tune.")
|
|
|
|
"Filepath of manifest to tune.")
|
|
|
|
add_arg('mean_std_path', str,
|
|
|
|
add_arg('mean_std_path', str,
|
|
|
|
'data/librispeech/mean_std.npz',
|
|
|
|
'data/librispeech/mean_std.npz',
|
|
|
@ -63,7 +67,7 @@ args = parser.parse_args()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def tune():
|
|
|
|
def tune():
|
|
|
|
"""Tune parameters alpha and beta on one minibatch."""
|
|
|
|
"""Tune parameters alpha and beta incrementally."""
|
|
|
|
if not args.num_alphas >= 0:
|
|
|
|
if not args.num_alphas >= 0:
|
|
|
|
raise ValueError("num_alphas must be non-negative!")
|
|
|
|
raise ValueError("num_alphas must be non-negative!")
|
|
|
|
if not args.num_betas >= 0:
|
|
|
|
if not args.num_betas >= 0:
|
|
|
@ -77,7 +81,7 @@ def tune():
|
|
|
|
num_threads=1)
|
|
|
|
num_threads=1)
|
|
|
|
batch_reader = data_generator.batch_reader_creator(
|
|
|
|
batch_reader = data_generator.batch_reader_creator(
|
|
|
|
manifest_path=args.tune_manifest,
|
|
|
|
manifest_path=args.tune_manifest,
|
|
|
|
batch_size=args.num_samples,
|
|
|
|
batch_size=args.batch_size,
|
|
|
|
sortagrad=False,
|
|
|
|
sortagrad=False,
|
|
|
|
shuffle_method=None)
|
|
|
|
shuffle_method=None)
|
|
|
|
tune_data = batch_reader().next()
|
|
|
|
tune_data = batch_reader().next()
|
|
|
@ -95,30 +99,75 @@ def tune():
|
|
|
|
pretrained_model_path=args.model_path,
|
|
|
|
pretrained_model_path=args.model_path,
|
|
|
|
share_rnn_weights=args.share_rnn_weights)
|
|
|
|
share_rnn_weights=args.share_rnn_weights)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# decoders only accept string encoded in utf-8
|
|
|
|
|
|
|
|
vocab_list = [chars.encode("utf-8") for chars in data_generator.vocab_list]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
error_rate_func = cer if args.error_rate_type == 'cer' else wer
|
|
|
|
# create grid for search
|
|
|
|
# create grid for search
|
|
|
|
cand_alphas = np.linspace(args.alpha_from, args.alpha_to, args.num_alphas)
|
|
|
|
cand_alphas = np.linspace(args.alpha_from, args.alpha_to, args.num_alphas)
|
|
|
|
cand_betas = np.linspace(args.beta_from, args.beta_to, args.num_betas)
|
|
|
|
cand_betas = np.linspace(args.beta_from, args.beta_to, args.num_betas)
|
|
|
|
params_grid = [(alpha, beta) for alpha in cand_alphas
|
|
|
|
params_grid = [(alpha, beta) for alpha in cand_alphas
|
|
|
|
for beta in cand_betas]
|
|
|
|
for beta in cand_betas]
|
|
|
|
|
|
|
|
|
|
|
|
## tune parameters in loop
|
|
|
|
err_sum = [0.0 for i in xrange(len(params_grid))]
|
|
|
|
for alpha, beta in params_grid:
|
|
|
|
err_ave = [0.0 for i in xrange(len(params_grid))]
|
|
|
|
|
|
|
|
num_ins, cur_batch = 0, 0
|
|
|
|
|
|
|
|
## incremental tuning parameters over multiple batches
|
|
|
|
|
|
|
|
for infer_data in batch_reader():
|
|
|
|
|
|
|
|
if (args.num_batches >= 0) and (cur_batch >= args.num_batches):
|
|
|
|
|
|
|
|
break
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
target_transcripts = [
|
|
|
|
|
|
|
|
''.join([data_generator.vocab_list[token] for token in transcript])
|
|
|
|
|
|
|
|
for _, transcript in infer_data
|
|
|
|
|
|
|
|
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
num_ins += len(target_transcripts)
|
|
|
|
|
|
|
|
# grid search
|
|
|
|
|
|
|
|
for index, (alpha, beta) in enumerate(params_grid):
|
|
|
|
result_transcripts = ds2_model.infer_batch(
|
|
|
|
result_transcripts = ds2_model.infer_batch(
|
|
|
|
infer_data=tune_data,
|
|
|
|
infer_data=infer_data,
|
|
|
|
decoding_method='ctc_beam_search',
|
|
|
|
decoding_method='ctc_beam_search',
|
|
|
|
beam_alpha=alpha,
|
|
|
|
beam_alpha=alpha,
|
|
|
|
beam_beta=beta,
|
|
|
|
beam_beta=beta,
|
|
|
|
beam_size=args.beam_size,
|
|
|
|
beam_size=args.beam_size,
|
|
|
|
cutoff_prob=args.cutoff_prob,
|
|
|
|
cutoff_prob=args.cutoff_prob,
|
|
|
|
vocab_list=data_generator.vocab_list,
|
|
|
|
cutoff_top_n=args.cutoff_top_n,
|
|
|
|
|
|
|
|
vocab_list=vocab_list,
|
|
|
|
language_model_path=args.lang_model_path,
|
|
|
|
language_model_path=args.lang_model_path,
|
|
|
|
num_processes=args.num_proc_bsearch)
|
|
|
|
num_processes=args.num_proc_bsearch)
|
|
|
|
wer_sum, num_ins = 0.0, 0
|
|
|
|
|
|
|
|
for target, result in zip(target_transcripts, result_transcripts):
|
|
|
|
for target, result in zip(target_transcripts, result_transcripts):
|
|
|
|
wer_sum += wer(target, result)
|
|
|
|
err_sum[index] += error_rate_func(target, result)
|
|
|
|
num_ins += 1
|
|
|
|
err_ave[index] = err_sum[index] / num_ins
|
|
|
|
print("alpha = %f\tbeta = %f\tWER = %f" %
|
|
|
|
if index % 2 == 0:
|
|
|
|
(alpha, beta, wer_sum / num_ins))
|
|
|
|
sys.stdout.write('.')
|
|
|
|
|
|
|
|
sys.stdout.flush()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# output on-line tuning result at the end of current batch
|
|
|
|
|
|
|
|
err_ave_min = min(err_ave)
|
|
|
|
|
|
|
|
min_index = err_ave.index(err_ave_min)
|
|
|
|
|
|
|
|
print("\nBatch %d [%d/?], current opt (alpha, beta) = (%s, %s), "
|
|
|
|
|
|
|
|
" min [%s] = %f" %(cur_batch, num_ins,
|
|
|
|
|
|
|
|
"%.3f" % params_grid[min_index][0],
|
|
|
|
|
|
|
|
"%.3f" % params_grid[min_index][1],
|
|
|
|
|
|
|
|
args.error_rate_type, err_ave_min))
|
|
|
|
|
|
|
|
cur_batch += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# output WER/CER at every (alpha, beta)
|
|
|
|
|
|
|
|
print("\nFinal %s:\n" % args.error_rate_type)
|
|
|
|
|
|
|
|
for index in xrange(len(params_grid)):
|
|
|
|
|
|
|
|
print("(alpha, beta) = (%s, %s), [%s] = %f"
|
|
|
|
|
|
|
|
% ("%.3f" % params_grid[index][0], "%.3f" % params_grid[index][1],
|
|
|
|
|
|
|
|
args.error_rate_type, err_ave[index]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
err_ave_min = min(err_ave)
|
|
|
|
|
|
|
|
min_index = err_ave.index(err_ave_min)
|
|
|
|
|
|
|
|
print("\nFinish tuning on %d batches, final opt (alpha, beta) = (%s, %s)"
|
|
|
|
|
|
|
|
% (args.num_batches, "%.3f" % params_grid[min_index][0],
|
|
|
|
|
|
|
|
"%.3f" % params_grid[min_index][1]))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ds2_model.logger.info("finish inference")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def main():
|
|
|
|
def main():
|
|
|
|