sp is added to the model.
This commit is contained in:
yemaozi88
parent
b1b1942fa0
commit
41d4fa5ff9
Binary file not shown.
@ -378,17 +378,22 @@ def ipa2htk(ipa):
|
||||
return ''.join(htk_splitted)
|
||||
|
||||
|
||||
def performance_on_stimmen(stimmen_dir, hmmdefs):
|
||||
#hmmdefs = r'c:\OneDrive\Research\rug\experiments\acoustic_model\fame\htk\model_\hmm1\iter20\hmmdefs'
|
||||
#stimmen_dir = r'c:\OneDrive\Research\rug\experiments\acoustic_model\fame\htk\stimmen'
|
||||
def performance_on_stimmen(config_dir, stimmen_dir, hmmdefs):
|
||||
lattice_file = os.path.join(stimmen_dir, 'word_lattice.ltc')
|
||||
hvite_scp = os.path.join(stimmen_dir, 'hvite.scp')
|
||||
#fh.make_filelist(os.path.join(stimmen_dir, 'mfc'), hvite_scp, file_type='mfc')
|
||||
hresult_scp = os.path.join(stimmen_dir, 'hresult.scp')
|
||||
#fh.make_filelist(os.path.join(stimmen_dir, 'mfc'), hresult_scp, file_type='rec')
|
||||
lexicon_file = os.path.join(stimmen_dir, 'lexicon_recognition.dic')
|
||||
chtk = pyhtk.HTK(config_dir, fame_asr.phoneset_htk, lexicon_file)
|
||||
|
||||
# get feature_size from hmmdefs.
|
||||
with open(hmmdefs) as f:
|
||||
line = f.readline()
|
||||
line = f.readline().strip()
|
||||
feature_size = int(line.split(' ')[2])
|
||||
|
||||
chtk = pyhtk.HTK(config_dir, fame_asr.phoneset_htk, lexicon_file, feature_size)
|
||||
|
||||
result = chtk.recognition(
|
||||
lattice_file,
|
||||
hmmdefs,
|
||||
|
||||
@ -26,7 +26,7 @@ make_mlf = 0
|
||||
extract_features = 0
|
||||
flat_start = 0
|
||||
train_model_without_sp = 0
|
||||
add_sp = 0
|
||||
add_sp = 1
|
||||
train_model_with_sp = 0
|
||||
train_model_with_sp_align_mlf = 0
|
||||
train_triphone = 0
|
||||
@ -35,6 +35,9 @@ train_triphone = 0
|
||||
|
||||
# pre-defined values.
|
||||
dataset_list = ['devel', 'test', 'train']
|
||||
feature_size = 39
|
||||
improvement_threshold = 0.5
|
||||
|
||||
hmmdefs_name = 'hmmdefs'
|
||||
proto_name = 'proto'
|
||||
|
||||
@ -47,7 +50,9 @@ sil_hed = os.path.join(config_dir, 'sil.hed')
|
||||
prototype = os.path.join(config_dir, proto_name)
|
||||
|
||||
model_dir = os.path.join(default.htk_dir, 'model')
|
||||
|
||||
model0_dir = os.path.join(model_dir, 'hmm0')
|
||||
model1_dir = os.path.join(model_dir, 'hmm1')
|
||||
model1sp_dir = os.path.join(model_dir, 'hmm1sp')
|
||||
|
||||
# directories / files to be made.
|
||||
lexicon_dir = os.path.join(default.htk_dir, 'lexicon')
|
||||
@ -55,9 +60,6 @@ lexicon_htk_asr = os.path.join(lexicon_dir, 'lex.htk_asr')
|
||||
lexicon_htk_oov = os.path.join(lexicon_dir, 'lex.htk_oov')
|
||||
lexicon_htk = os.path.join(lexicon_dir, 'lex.htk')
|
||||
|
||||
|
||||
#model1_dir = os.path.join(model_dir, 'hmm1')
|
||||
|
||||
feature_dir = os.path.join(default.htk_dir, 'mfc')
|
||||
fh.make_new_directory(feature_dir, existing_dir='leave')
|
||||
tmp_dir = os.path.join(default.htk_dir, 'tmp')
|
||||
@ -65,11 +67,17 @@ fh.make_new_directory(tmp_dir, existing_dir='leave')
|
||||
label_dir = os.path.join(default.htk_dir, 'label')
|
||||
fh.make_new_directory(label_dir, existing_dir='leave')
|
||||
|
||||
|
||||
|
||||
## training
|
||||
hcompv_scp_train = os.path.join(tmp_dir, 'train.scp')
|
||||
mlf_file_train = os.path.join(label_dir, 'train_phone.mlf')
|
||||
mlf_file_train_aligned = os.path.join(label_dir, 'train_phone_aligned.mlf')
|
||||
|
||||
## testing
|
||||
htk_stimmen_dir = os.path.join(default.htk_dir, 'stimmen')
|
||||
|
||||
|
||||
## train without sp
|
||||
niter_max = 10
|
||||
|
||||
@ -100,7 +108,7 @@ if make_lexicon:
|
||||
|
||||
|
||||
## intialize the instance for HTK.
|
||||
chtk = pyhtk.HTK(config_dir, fame_asr.phoneset_htk, lexicon_htk)
|
||||
chtk = pyhtk.HTK(config_dir, fame_asr.phoneset_htk, lexicon_htk, feature_size)
|
||||
|
||||
|
||||
## ======================= make label files =======================
|
||||
@ -223,11 +231,14 @@ if extract_features:
|
||||
if flat_start:
|
||||
timer_start = time.time()
|
||||
print('==== flat start ====')
|
||||
feature_size = 39
|
||||
model0_dir = os.path.join(model_dir, 'hmm0')
|
||||
fh.make_new_directory(model0_dir, existing_dir='leave')
|
||||
|
||||
chtk.flat_start(hcompv_scp_train, model0_dir, feature_size)
|
||||
chtk.flat_start(hcompv_scp_train, model0_dir)
|
||||
|
||||
# create macros.
|
||||
vFloors = os.path.join(model0_dir, 'vFloors')
|
||||
if os.path.exists(vFloors):
|
||||
chtk.create_macros(vFloors)
|
||||
|
||||
# allocate mean & variance to all phones in the phone list
|
||||
print('>>> allocating mean & variance to all phones in the phone list...')
|
||||
@ -241,69 +252,38 @@ if flat_start:
|
||||
|
||||
## ======================= train model without short pause =======================
|
||||
if train_model_without_sp:
|
||||
fh.make_new_directory(model1_dir)
|
||||
|
||||
print('==== train model without sp ====')
|
||||
if not os.path.exists(os.path.join(model1_dir, 'iter0')):
|
||||
shutil.copytree(model0_dir, os.path.join(model1_dir, 'iter0'))
|
||||
for niter in range(1, niter_max):
|
||||
timer_start = time.time()
|
||||
hmm_n = 'iter' + str(niter)
|
||||
hmm_n_pre = 'iter' + str(niter-1)
|
||||
modeln_dir = os.path.join(model1_dir, hmm_n)
|
||||
modeln_dir_pre = os.path.join(model1_dir, hmm_n_pre)
|
||||
|
||||
# re-estimation
|
||||
fh.make_new_directory(modeln_dir)
|
||||
pyhtk.re_estimation(
|
||||
config_train,
|
||||
os.path.join(modeln_dir_pre, hmmdefs_name),
|
||||
modeln_dir,
|
||||
hcompv_scp_train, phonelist_txt,
|
||||
mlf_file=mlf_file_train,
|
||||
macros=os.path.join(modeln_dir_pre, 'macros'))
|
||||
print("elapsed time: {}".format(time.time() - timer_start))
|
||||
|
||||
timer_start = time.time()
|
||||
niter = chtk.re_estimation_until_saturated(
|
||||
model1_dir,
|
||||
model0_dir, improvement_threshold, hcompv_scp_train,
|
||||
os.path.join(htk_stimmen_dir, 'mfc'),
|
||||
'mfc',
|
||||
os.path.join(htk_stimmen_dir, 'word_lattice.ltc'),
|
||||
mlf_file=mlf_file_train,
|
||||
lexicon_file=os.path.join(htk_stimmen_dir, 'lexicon_recognition.dic')
|
||||
)
|
||||
|
||||
print("elapsed time: {}".format(time.time() - timer_start))
|
||||
|
||||
|
||||
## ======================= adding sp to the model =======================
|
||||
if add_sp:
|
||||
print('==== adding sp to the model ====')
|
||||
# reference:
|
||||
# http://www.f.waseda.jp/yusukekondo/htk.html#flat_start_estimation
|
||||
|
||||
# make model with sp.
|
||||
print('>>> modifying the last model in the previous step...')
|
||||
modeln_dir_pre = os.path.join(model1_dir, 'iter'+str(niter_max-1))
|
||||
modeln_dir = modeln_dir_pre.replace('iter' + str(niter_max-1), 'iter' + str(niter_max))
|
||||
fh.make_new_directory(modeln_dir)
|
||||
shutil.copy(
|
||||
os.path.join(modeln_dir_pre, 'macros'),
|
||||
os.path.join(modeln_dir, 'macros'))
|
||||
shutil.copy(
|
||||
os.path.join(modeln_dir_pre, hmmdefs_name),
|
||||
os.path.join(modeln_dir, hmmdefs_name))
|
||||
niter = 7
|
||||
print('>>> adding sp state to the last model in the previous step...')
|
||||
fh.make_new_directory(model1sp_dir, existing_dir='leave')
|
||||
modeln_dir_pre = os.path.join(model1_dir, 'iter'+str(niter))
|
||||
|
||||
## =======================
|
||||
## manually make changes to modeln_dir/hmmdefs
|
||||
## =======================
|
||||
# add states 'sil'.
|
||||
# http://www.f.waseda.jp/yusukekondo/htk.html#flat_start_estimation
|
||||
#shutil.copy(
|
||||
# os.path.join(model_dir, 'hmmdefs.txt'),
|
||||
# os.path.join(modeln_dir, hmmdefs_name))
|
||||
|
||||
#hmmdefs_file_pre = os.path.join(modeln_dir_pre, hmmdefs_name)
|
||||
hmmdefs_file = os.path.join(modeln_dir, hmmdefs_name)
|
||||
macros_file = os.path.join(modeln_dir, 'macros')
|
||||
#with open(hmmdefs_file_pre) as f:
|
||||
# lines = f.read()
|
||||
#lines_ = lines.split('~h ')
|
||||
#sil_model = [line for line in lines_ if line.split('\n')[0].replace('"', '') == 'sil'][0]
|
||||
|
||||
# update hmmdefs and macros.
|
||||
print('>>> updating hmmdefs and macros...')
|
||||
modeln_dir_pre = modeln_dir
|
||||
modeln_dir = modeln_dir.replace('iter' + str(niter_max), 'iter' + str(niter_max+1))
|
||||
fh.make_new_directory(modeln_dir)
|
||||
pyhtk.include_sil_in_hmmdefs(macros_file, hmmdefs_file, modeln_dir, sil_hed, phonelist_txt)
|
||||
## update hmmdefs and macros.
|
||||
print('>>> adding sp to the model...')
|
||||
modeln_dir = os.path.join(model1sp_dir, 'iter0')
|
||||
chtk.add_sp(modeln_dir_pre, modeln_dir)
|
||||
|
||||
|
||||
## ======================= train model with short pause =======================
|
||||
|
||||
Loading…
Reference in New Issue
Block a user