def classify_htk(audio_file):
MIN_BREATH_DUR = 0.1
MIN_AC = 500
cwd = os.getcwd()
os.chdir(os.path.dirname(os.path.realpath(__file__)))
transcript = "breath.transcript"
output = "breath-classify-output.json"
results = "tmp/aligned.results"
with open(transcript, 'w') as f:
f.write("""[{"speaker": "speaker", "line": "{BR}"}]""")
do_alignment(audio_file, transcript, output, json=True, textgrid=False)
os.remove(transcript)
# subprocess.call('python ../p2fa/align.py ../%s %s %s' %
# (audio_file, transcript, output), shell=True)
final_words = [{
"start": 0.0,
"end": 0.0,
"alignedWord": "sp",
"word": "{p}"
}]
with open(results, 'r') as res:
match = ac_re.search(res.read())
if match:
ac = int(match.group(1))
print "Ac:", ac
if ac > MIN_AC:
print "Breath!"
with open(output, 'r') as out:
words = json.load(out)["words"]
breath = filter(
lambda x: x["alignedWord"] == "{BR}",
words)[0]
breath_dur = breath["end"] - breath["start"]
print "breath len", breath_dur
if breath_dur > MIN_BREATH_DUR:
final_words = words
final_words[0]["start"] = 0.0
for word in final_words:
if word["alignedWord"] == "{BR}":
word["likelihood"] = ac
os.chdir(cwd)
return final_words
def align_transcripts(as_ep_ids):
"""
Align the transcript segments in ./alignment_data/seg_json/ to episode
audio using p2fa. If there are any errors during alignment, skip the
episode on which they occur and move on.
Parameters
----------
as_ep_ids : list
The list of all audiosearch episode ids associated with a particular
podcast
Returns
-------
as_ep_ids : list
A list of episode ids that have been completely aligned.
problem_episodes : list
A list of episode ids that were not fully aligned
"""
problem_episodes = []
for ep_num in as_ep_ids:
wavfile = './alignment_data/seg_audio/{}_seg*.wav'.format(ep_num)
for ff in glob.glob(wavfile):
file_name = os.path.split(ff)[-1].split('.')[0]
trsfile = './alignment_data/seg_json/{}.json'.format(file_name)
outfile = './alignment_data/alignments_json/' \
'{}_aligned.json'.format(file_name)
if os.path.lexists(outfile):
fn = os.path.split(outfile)[-1]
print('\t{} already exists! Skipping'.format(fn))
continue
try:
print('\n\tAligning {}...'.format(file_name))
align.do_alignment(ff,
trsfile,
outfile,
json=True,
textgrid=False,
phonemes=True,
breaths=False)
except:
print('\n\tError Aligning {}'.format(file_name))
problem_episodes.append(ep_num)
as_ep_ids.remove(ep_num)
continue
return as_ep_ids, problem_episodes
def align_transcripts(as_ep_ids):
"""
Align the transcript segments in ./alignment_data/seg_json/ to episode
audio using p2fa. If there are any errors during alignment, skip the
episode on which they occur and move on.
Parameters
----------
as_ep_ids : list
The list of all audiosearch episode ids associated with a particular
podcast
Returns
-------
as_ep_ids : list
A list of episode ids that have been completely aligned.
problem_episodes : list
A list of episode ids that were not fully aligned
"""
problem_episodes = []
for ep_num in as_ep_ids:
wavfile = './alignment_data/seg_audio/{}_seg*.wav'.format(ep_num)
for ff in glob.glob(wavfile):
file_name = os.path.split(ff)[-1].split('.')[0]
trsfile = './alignment_data/seg_json/{}.json'.format(file_name)
outfile = './alignment_data/alignments_json/' \
'{}_aligned.json'.format(file_name)
if os.path.lexists(outfile):
fn = os.path.split(outfile)[-1]
print('\t{} already exists! Skipping'.format(fn))
continue
try:
print('\n\tAligning {}...'.format(file_name))
align.do_alignment(ff, trsfile, outfile, json=True,
textgrid=False, phonemes=True,
breaths=False)
except:
print('\n\tError Aligning {}'.format(file_name))
problem_episodes.append(ep_num)
as_ep_ids.remove(ep_num)
continue
return as_ep_ids, problem_episodes
