onyx.htkfiles.htkscoring – Quick test of scoring some acoustic models with feature data.

>>> module_dir, module_name = os.path.split(__file__)

>>> am_filename = os.path.join(module_dir, "monophones.mmf")
>>> with open(am_filename) as f:
...     models_dict, hmm_mgr, gmm_mgr = read_htk_mmf_file(f)
>>> print ' '.join(sorted(models_dict))
aa ae ah ao aw ax ay b ch d dd dh dx eh el en er ey f g hh ih iy jh k kd l m n ng ow oy p pd r s sh sil sp t td th ts uh uw v w y z

>>> print hmm_mgr
HmmMgr with 49 models
>>> hmm_mgr.num_models
49
>>> hmm_mgr.get_adaptation_state()
'NOT_ADAPTING'

>>> print gmm_mgr
GmmMgr with 145 models of type <class 'onyx.am.gaussian.GaussianMixtureModel'>
  (dimension 39, covariance_type onyx.am.gaussian.GaussianModelBase.DIAGONAL_COVARIANCE)
>>> gmm_mgr.num_models
145
>>> gmm_mgr.get_adaptation_state()
'NOT_ADAPTING'

>>> audio_filename = onyx.home + '/data/htk_r1_adg0_4/' + "adg0_4_sr009.mfc"
>>> with open(audio_filename, 'rb') as f:
...    (audio_data, (kind, qualifiers), samp_period) = read_htk_audio_file(f)
>>> kind
(6, 'MFCC', 'mel-frequency cepstral coefficients')
>>> qualifiers
(('A', 'has acceleration coefficients'), ('E', 'has energy'), ('D', 'has delta coefficients'))

>>> audio_data.shape
(449, 39)
>>> len(audio_data)
449
>>> models = tuple(gmm_mgr.get_model(i) for i in xrange(gmm_mgr.num_models))
>>> global_stats = simple_stats()
>>> for obs_id, observation in enumerate(audio_data):
...   #obs_stats = simple_stats()
...   for model in models:
...     score = model.score(observation)
...     #obs_stats(score)
...     global_stats(score)
...   #print ' ', obs_id, 'obs_stats:',  obs_stats

>>> print 'global_stats:', global_stats 
global_stats: count 65105  min +(-0918)0x7b58d14eca40...  max +(-0072)0xc1d3354e8e44...  sum +(-0070)0x91f1fafccc92...  sumsq +(-0142)0x927de65a98b...
>>> global_stats.count == len(audio_data) * gmm_mgr.num_models
True

Same test but with adapted models

>>> am_filename = os.path.join(module_dir, "monophones4.mmf")
>>> with open(am_filename) as f:
...     models_dict, hmm_mgr, gmm_mgr = read_htk_mmf_file(f)
>>> print ' '.join(sorted(models_dict))
aa ae ah ao aw ax ay b ch d dd dh dx eh el en er ey f g hh ih iy jh k kd l m n ng ow oy p pd r s sh sil sp t td th ts uh uw v w y z

>>> models = tuple(gmm_mgr.get_model(i) for i in xrange(gmm_mgr.num_models))
>>> global_stats = simple_stats()
>>> for obs_id, observation in enumerate(audio_data):
...   for model in models:
...     score = model.score(observation)
...     global_stats(score)

Three of the stats differ in the LSB on 64-bit platform.

>>> print 'global_stats:', global_stats 
global_stats: count 65105  min +(-1000)0xbadbadbadbad0  max +(-0061)0xbf7e129869...  sum +(-0060)0x8e9d15018691...  sumsq +(-0121)0xcd929a6bbdb...
>>> global_stats.count == len(audio_data) * gmm_mgr.num_models
True

class onyx.htkfiles.htkscoring.simple_stats

Bases: object