def compare_pronunciations(word, show_name, file_name):
"""
Generates a plot showing the frequency of each pronunciation of a given
word in corpus.
"""
word = word.lower()
as_client = init_as_client()
as_ep_ids, as_show_id = find_episodes(show_name, as_client)
pdict_fp = './alignment_data/pronunciations/' \
'{}_prnc_dict.pickle'.format(file_name)
if os.path.lexists(pdict_fp):
with open(pdict_fp, 'rb') as handle:
prnc_dict = pickle.load(handle)
else:
prnc_dict = compile_prnc_dict(as_ep_ids, file_name)
plot_pronunciations(prnc_dict, word, show_name, file_name)
return prnc_dict
def align_show(show_name, n_segs, file_name):
"""
Finds all the episode ids associated with show_name in the audiosearch db,
find the corresponding transcript ids for each episode, downloads the
episode audio from soundcloud, compiles the episode transcript from the
audiosearch db, segments the audio and transcripts into bite-sized
segments, runs each segment through p2fa to get phoneme-level
alignments, and writes both regular and phoneme-level bookworm files.
Parameters
----------
show_name : str
The name of the particular podcast as it appears in the audiosearch API
n_segs : int
The number of chunks to split a transcript into during alignment
with p2fa (more chunks ~> faster alignment)
file_name : str
A nickname for the podcast. This is used when saving the audio files,
naming the bookworm folders, writing transcript jsons, etc.
"""
make_alignments_directory()
as_client = init_as_client()
as_ep_ids, as_show_id = find_episodes(show_name, as_client)
trans_dict = find_episode_transcript_ids(as_ep_ids)
compile_audio_and_transcripts(trans_dict, n_segs, as_client, file_name)
as_ep_ids, problem_episodes = align_transcripts(as_ep_ids)
compile_alignments_bookworm(as_ep_ids, file_name)
prnc_dict = compile_prnc_dict(as_ep_ids, pdict_fp)
pause_dict = compile_pause_dict(file_name, prnc_dict)
plot_pause_words(show_name, file_name, pause_dict)
def align_show(show_name, n_segs, file_name):
"""
Finds all the episode ids associated with show_name in the audiosearch db,
find the corresponding transcript ids for each episode, downloads the
episode audio from soundcloud, compiles the episode transcript from the
audiosearch db, segments the audio and transcripts into bite-sized
segments, runs each segment through p2fa to get phoneme-level
alignments, and writes both regular and phoneme-level bookworm files.
Parameters
----------
show_name : str
The name of the particular podcast as it appears in the audiosearch API
n_segs : int
The number of chunks to split a transcript into during alignment
with p2fa (more chunks ~> faster alignment)
file_name : str
A nickname for the podcast. This is used when saving the audio files,
naming the bookworm folders, writing transcript jsons, etc.
"""
make_alignments_directory()
as_client = init_as_client()
as_ep_ids, as_show_id = find_episodes(show_name, as_client)
trans_dict = find_episode_transcript_ids(as_ep_ids)
compile_audio_and_transcripts(trans_dict, n_segs, as_client, file_name)
as_ep_ids, problem_episodes = align_transcripts(as_ep_ids)
compile_alignments_bookworm(as_ep_ids, file_name)
prnc_dict = compile_prnc_dict(as_ep_ids, pdict_fp)
pause_dict = compile_pause_dict(file_name, prnc_dict)
plot_pause_words(show_name, file_name, pause_dict)
def compile_alignments_bookworm(as_ep_ids, file_name):
"""
Reads the phoneme alignments for each episode, matches them to
the corresponding line in transcript, gets the appropriate episode
metadata, and constructs the corresponding catalog.json and input.txt
files for both a phoneme bookworm and a regular bookworm.
Parameters
----------
as_ep_ids : list
The list of audiosearch episode ids associated with a particular
podcast
file_name : str
A nickname for the podcast. This is used when saving the audio files,
naming the bookworm folders, writing transcript jsons, etc.
"""
counter = 0
# db = db_connect()
db = Postgres_Connect().connection
as_client = init_as_client()
make_bw_directories(file_name)
ts_csv = load_timesteps(file_name)
trans_dict = find_episode_transcript_ids(as_ep_ids)
for (ep_num, trans_id) in trans_dict.items():
phoneme_file = \
'./alignment_data/alignments_json/{}_seg*.json'.format(ep_num)
if len(glob.glob(phoneme_file)) == 0:
print('No phoneme alignments for episode' \
' {}. Skipping.'.format(ep_num))
continue
transcript = compile_episode_transcript(trans_id, db)
meta = collect_episode_metadata(db, ep_num, as_client)
if len(transcript) == 0:
print('Unable to find transcript ID {} ' \
'in AS database. Skipping.'.format(trans_id))
continue
phoneme_transcript = []
for fp in np.sort(glob.glob(phoneme_file)):
phoneme_transcript = phoneme_transcript + \
compile_phoneme_transcript(fp, transcript)
print('Writing BW entry for {}'.format(phoneme_file))
# phoneme_transcript[idx] =
# [phoneme sentence, transcript sentence, transcript index]
phoneme_transcript = np.asarray(phoneme_transcript)
counter = write_bw_catalog(transcript, phoneme_transcript,
counter, ep_num, meta, file_name)
write_field_descriptions(file_name)
shutil.move('./metadata', './' + file_name)
shutil.move('./texts', './' + file_name)
os.mkdir('./' + file_name + '_bookworm')
shutil.move('./' + file_name, './' + file_name + '_bookworm')
shutil.move('./' + file_name + '_phonemes', './' + file_name + '_bookworm')
def compile_alignments_bookworm(as_ep_ids, file_name):
"""
Reads the phoneme alignments for each episode, matches them to
the corresponding line in transcript, gets the appropriate episode
metadata, and constructs the corresponding catalog.json and input.txt
files for both a phoneme bookworm and a regular bookworm.
Parameters
----------
as_ep_ids : list
The list of audiosearch episode ids associated with a particular
podcast
file_name : str
A nickname for the podcast. This is used when saving the audio files,
naming the bookworm folders, writing transcript jsons, etc.
"""
counter = 0
# db = db_connect()
db = Postgres_Connect().connection
as_client = init_as_client()
make_bw_directories(file_name)
ts_csv = load_timesteps(file_name)
trans_dict = find_episode_transcript_ids(as_ep_ids)
for (ep_num, trans_id) in trans_dict.items():
phoneme_file = \
'./alignment_data/alignments_json/{}_seg*.json'.format(ep_num)
if len(glob.glob(phoneme_file)) == 0:
print('No phoneme alignments for episode' \
' {}. Skipping.'.format(ep_num))
continue
transcript = compile_episode_transcript(trans_id, db)
meta = collect_episode_metadata(db, ep_num, as_client)
if len(transcript) == 0:
print('Unable to find transcript ID {} ' \
'in AS database. Skipping.'.format(trans_id))
continue
phoneme_transcript = []
for fp in np.sort(glob.glob(phoneme_file)):
phoneme_transcript = phoneme_transcript + \
compile_phoneme_transcript(fp, transcript)
print('Writing BW entry for {}'.format(phoneme_file))
# phoneme_transcript[idx] =
# [phoneme sentence, transcript sentence, transcript index]
phoneme_transcript = np.asarray(phoneme_transcript)
counter = write_bw_catalog(transcript, phoneme_transcript, counter,
ep_num, meta, file_name)
write_field_descriptions(file_name)
shutil.move('./metadata', './' + file_name)
shutil.move('./texts', './' + file_name)
os.mkdir('./' + file_name + '_bookworm')
shutil.move('./' + file_name, './' + file_name + '_bookworm')
shutil.move('./' + file_name + '_phonemes', './' + file_name + '_bookworm')