def generate_signal():
rng = np.random.RandomState(seed = 1024)
s = (2**3)*rng.randn(fs*10).astype(float_cpu(), copy=False)
vad = np.zeros((len(s),), dtype=bool)
vad[2*fs:8*fs] = True
s += (2**12)*vad.astype(dtype=float_cpu())*np.sign(s)
vad = vad[::160]
#s = rng.randn(fs*10).astype(float_cpu(), copy=False)
return s, vad
def generate_features():
rng = np.random.RandomState(seed=1024)
x = rng.randn(10, 2).astype(float_cpu(), copy=False)
vad = np.zeros((10, ), dtype='bool')
vad[4:8] = 1
return x, vad
def generate_features():
rng = np.random.RandomState(seed=1024)
x = rng.randn(60 * 100, 2).astype(float_cpu(), copy=False)
x *= rng.rand(60 * 100, 1)
return x
def to_sparse(r, num_comp):
index = np.argsort(r, axis=1)[:, -num_comp:]
r_sparse = np.zeros((r.shape[0], num_comp), dtype=float_cpu())
for i, index_i in enumerate(index):
r_sparse[i] = r[i, index_i]
r_sparse = r_sparse / np.sum(r_sparse, axis=-1, keepdims=True)
return r_sparse, index
def combine_diar_scores(ndx, orig_seg, subseg_scores):
scores = np.zeros(ndx.trial_mask.shape, dtype=float_cpu())
for j in xrange(len(ndx.seg_set)):
idx = orig_seg == ndx.seg_set[j]
subseg_scores_j = subseg_scores[:, idx]
scores_j = np.max(subseg_scores_j, axis=1)
scores[:,j] = scores_j
return scores
def gen_signals(num_signals=3):
rng = np.random.RandomState(seed=1)
s = []
keys = []
for i in xrange(num_signals):
s_i = rng.randn(fs)
s_i = ((2**15 - 1) / np.max(np.abs(s_i)) * s_i).astype('int32').astype(
float_cpu())
s.append(s_i)
keys.append('s%d' % i)
return keys, s
def eval_elbo(seq_file, file_list, model_file, preproc_file, output_file,
ubm_type, **kwargs):
sr_args = SR.filter_eval_args(**kwargs)
if preproc_file is not None:
preproc = TransformList.load(preproc_file)
else:
preproc = None
sr = SR(seq_file,
file_list,
batch_size=1,
shuffle_seqs=False,
preproc=preproc,
**sr_args)
t1 = time.time()
if ubm_type == 'diag-gmm':
model = DiagGMM.load(model_file)
else:
model = DiagGMM.load_from_kaldi(model_file)
model.initialize()
elbo = np.zeros((sr.num_seqs, ), dtype=float_cpu())
num_frames = np.zeros((sr.num_seqs, ), dtype=int)
keys = []
for i in xrange(sr.num_seqs):
x, key = sr.read_next_seq()
keys.append(key)
elbo[i] = model.elbo(x)
num_frames[i] = x.shape[0]
num_total_frames = np.sum(num_frames)
total_elbo = np.sum(elbo)
total_elbo_norm = total_elbo / num_total_frames
logging.info('Extract elapsed time: %.2f' % (time.time() - t1))
s = 'Total ELBO: %f\nELBO_NORM %f' % (total_elbo, total_elbo_norm)
logging.info(s)
with open(output_file, 'w') as f:
f.write(s)
def generate_signal():
rng = np.random.RandomState(seed=1024)
s = (2**10) * rng.randn(fs * 10).astype(float_cpu(), copy=False)
#s = rng.randn(fs*10).astype(float_cpu(), copy=False)
return s
def to_dense(r_sparse, index, num_comp):
r = np.zeros((r_sparse.shape[0], num_comp), dtype=float_cpu())
for i in xrange(r_sparse.shape[0]):
r[i, index[i]] = r_sparse[i]
return r
