mimick_segments.py

# When an audio has the same data in left and right channel, transcript end up looking like an echo between speakers
# in and out. The following code will filter out one of the speaker, and produce a text file with only one speaker's
# transcript. Because of the overtalk, segments are as short as one word. To mimick real segment, it concatenates
# words within a segment as long as there is less than 300ms between the end of a word and the beginning of the next one.
import json

# given a JSON file containing a call from Activate API. https://docs.allo-media.net/activate-api/

with open('call.json') as jsonfile:
  call = json.load(jsonfile)
  
unique_id = call['unique_id']
transcript = call['transcript_json']['callData']
# keep only IN speaker
transcript = [segment for segment in transcript if segment['from'] == 'in']
lines = []
timecode = 0
tmp_line = []
for segment in transcript:
  if timecode + 0.3 < segment['datetime']:
    lines.append(" ".join(tmp_line))
    tmp_line = []
  tmp_line.append(segment['content'])
  timecode = segment['datetime'] + segment['duration']
with open(f"{original_id}.txt", "w") as txtfile:
  txtfile.write("\n".join(lines))