More timedeltas (dmarx#89)

VERSION

@@ -1 +1 @@
-0.1.5
+0.1.6

vktrs/asr.py

@@ -5,13 +5,13 @@
   pip install git+https://github.com/openai/whisper
 """
 
-import datetime as dt
 import time
 
 import tokenizations
 from vktrs.utils import remove_punctuation
 import whisper
 
+
 def whisper_transcribe(
     audio_fpath="audio.mp3",
 ):
@@ -175,7 +175,6 @@ def whisper_segment_transcription(
     return prompt_starts
 
 
-
 def whisper_lyrics(audio_fpath="audio.mp3"):
     whispers = whisper_transcribe(audio_fpath)
     tiny2large, large2tiny, whispers_tokens = whisper_align(whispers)

vktrs/youtube.py

@@ -53,7 +53,7 @@ def parse_timestamp(ts):
         minutes=t.minute,
         seconds=t.second,
         microseconds=t.microsecond,
-        )
+        ).total_seconds()
 
 def vtt_to_token_timestamps(captions):
 
@@ -76,9 +76,9 @@ def vtt_to_token_timestamps(captions):
     token_start_times = []
     for line in all_word_starts_raw:
       starts_ = [
-          {'ts':hit[1], 
+          {'ts_str':hit[1], 
            'tok':hit[3].strip(),
-           'td':parse_timestamp(hit[1])
+           'ts':parse_timestamp(hit[1])
            } 
            for hit in re.findall(pat, line)]
       token_start_times.extend(starts_)
@@ -89,10 +89,9 @@ def srv2_to_token_timestamps(srv2_xml):
     srv2_soup = BeautifulSoup(srv2_xml, 'xml')
     return [
         {
-         'ts':e['t'], 
+         'ts_str':e['t'], 
          'tok':e.text,
-         #'td':dt.timedelta(microseconds=int(e['t']))
-         'td':dt.timedelta(milliseconds=int(e['t']))
+         'ts':dt.timedelta(milliseconds=int(e['t'])).total_seconds()
          } 
         for e in srv2_soup.find_all('text')
         if e.text.strip()