def position_scan_regions(sig, detect_mask, scan_len, detect_scan_olap):
detect_segments = segs_list_from_signal(detect_mask)
assert (utils_sig.is_array(sig))
assert (utils_sig.is_array(detect_mask))
assert ((scan_len % 2) == 1)
half_seg_len = (scan_len - 1) / 2
fsig = condition_signal_for_emph_scanning(sig)
res = []
for dseg in detect_segments:
dchunk = fsig[dseg['st']:dseg['end']]
max_idx = numpy.argmax(dchunk)
#assert(len(max_idx) == 1) # check that we have only 1 maximum value in the result, otherwise it is VERY strange
scan_set_st = int(dseg['st'] + max_idx - half_seg_len)
scan_set_end = int(dseg['st'] + max_idx + half_seg_len)
scan_seg = {'st': scan_set_st, 'end': scan_set_end}
# make sure that detection segment and corresponding scan-segment overlap on at least 50%
if get_seg_overlap(dseg, scan_seg) < detect_scan_olap:
scan_seg = adjust_seg_overlap(dseg, scan_seg, len(sig),
detect_scan_olap)
if scan_seg is not None:
res.append(scan_seg)
return res
def update_detection_results(mask, samplerate, detect_hysteresis,
merge_threshold, min_len, max_len):
assert (utils_sig.is_array(mask))
hyst_step = int(numpy.round(detect_hysteresis * samplerate))
merge_step = int(numpy.round(merge_threshold * samplerate))
min_keep_len = int(numpy.round(min_len * samplerate))
max_keep_len = int(numpy.round(max_len * samplerate))
# MY_DBG
#print("hyst_step = {0} merge_step = {1} ...".format(hyst_step, merge_step))
sig_len = len(mask)
idx = 0
result = numpy.copy(mask)
while (idx < sig_len):
if (mask[idx] > 0.0):
result[idx:idx + hyst_step] = 1.0
idx += hyst_step
continue
idx += 1
detect_segs = segs_list_from_signal(result)
detect_segs = merge_segs(detect_segs,
min_len=min_keep_len,
max_len=max_keep_len,
merge_thr=merge_step)
result = segs_list_to_signal(detect_segs, len(result))
return result
def remove_silence(sig, sil_thr=0.001):
assert (utils_sig.is_vector(sig) or utils_sig.is_array(sig))
tmp = sig.squeeze()
tmp = tmp[numpy.abs(tmp) >= sil_thr]
return utils_sig.preserve_shape(tmp, sig)
def plot_emphasis_scan_segs(signal,
detect,
scan_segs,
samplerate,
filename=None):
assert (utils_sig.is_array(signal))
assert (utils_sig.is_array(detect))
sig_time = numpy.arange(len(signal)) / samplerate
sig_cond = utils_emph.condition_signal_for_emph_scanning(signal)
plt_y = [signal, sig_cond, detect]
plt_x = [sig_time, sig_time, sig_time]
(seg_x, seg_y) = get_scan_seg_plots(scan_segs, samplerate, len(signal))
plt_y = plt_y + seg_y
plt_x = plt_x + seg_x
return plot_curves(plt_y, plt_x, saveto=filename)
def segs_list_from_signal(sig):
idx = 0
assert (utils_sig.is_array(sig))
res = []
while (idx < len(sig)):
if (sig[idx] > 0):
st = idx
while (sig[idx] > 0):
idx += 1
end = idx
res.append({'st': st, 'end': end})
continue
idx += 1
return res
def perform_mvn_norm(sig, skip_zeros=False):
assert utils_sig.is_vector(sig) or utils_sig.is_array(sig)
uv_mask = (sig == 0.0)
if skip_zeros:
tmp_sig = remove_silence(sig, 0.0)
mean = numpy.mean(tmp_sig)
std = numpy.std(tmp_sig)
result = (sig - mean) / std
result[uv_mask] = 0.0
return result
else:
mean = numpy.mean(sig)
std = numpy.std(sig)
result = (sig - mean) / std
result[uv_mask] = 0.0
return result