malarinv
/
jasper-asr
mirror of https://github.com/malarinv/jasper-asr.git
1
0
Fork 0
Code Issues Releases Wiki Activity

fix skipping null audio and add more verbose logs

Malar Kannan 2020-05-27 15:49:58 +05:30
parent a38789d0c3
commit 6f395af10d
2 changed files with 12 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
jasper/data/asr_recycler.py
View File

@ -59,10 +59,13 @@ def extract_data(
code_wav = code_fb.getvalue() code_wav = code_fb.getvalue()
# only starting 1 min audio has reliable alignment ignore rest # only starting 1 min audio has reliable alignment ignore rest
if start_time > 60: if start_time > 60:
print(f'start time over 60 seconds of audio skipping.')
break break
# only if some reasonable audio data is present yield it # only if some reasonable audio data is present yield it
if code_seg.duration_seconds >= 0.5: if code_seg.duration_seconds < 0.5:
yield transcript, code_seg.duration_seconds, code_wav print(f'transcript chunk "{transcript}" contains no audio skipping.')
continue
yield transcript, code_seg.duration_seconds, code_wav
def asr_data_generator(call_wav, call_wav_fname, events): def asr_data_generator(call_wav, call_wav_fname, events):
call_wav_0, call_wav_1 = call_wav.split_to_mono() call_wav_0, call_wav_1 = call_wav.split_to_mono()

9
jasper/data/rev_recycler.py
View File

@ -91,8 +91,10 @@ def extract_data(
text = "".join(lens["elements"].Each()["value"].collect()(monologue)) text = "".join(lens["elements"].Each()["value"].collect()(monologue))
text_clean = re.sub(r"\[.*\]", "", text) text_clean = re.sub(r"\[.*\]", "", text)
# only if some reasonable audio data is present yield it # only if some reasonable audio data is present yield it
if tscript_wav_seg.duration_seconds >= 0.5: if tscript_wav_seg.duration_seconds < 0.5:
yield text_clean, tscript_wav_seg.duration_seconds, tscript_wav print(f'transcript chunk "{text_clean}" contains no audio in {wav_path} skipping.')
continue
yield text_clean, tscript_wav_seg.duration_seconds, tscript_wav
def mono_asr_data_generator(wav_seg, wav_path, meta): def mono_asr_data_generator(wav_seg, wav_path, meta):
monologues = lens["monologues"].Each().collect()(meta) monologues = lens["monologues"].Each().collect()(meta)
@ -122,6 +124,9 @@ def extract_data(
tscript_wav = tscript_wav_fb.getvalue() tscript_wav = tscript_wav_fb.getvalue()
text = "".join(lens["elements"].Each()["value"].collect()(monologue)) text = "".join(lens["elements"].Each()["value"].collect()(monologue))
text_clean = re.sub(r"\[.*\]", "", text) text_clean = re.sub(r"\[.*\]", "", text)
if tscript_wav_seg.duration_seconds < 0.5:
print(f'transcript chunk "{text_clean}" contains no audio in {wav_path} skipping.')
continue
yield text_clean, tscript_wav_seg.duration_seconds, tscript_wav yield text_clean, tscript_wav_seg.duration_seconds, tscript_wav
def generate_rev_asr_data(): def generate_rev_asr_data():