transcribe.py

import json
import time
import whisper

MODEL_TYPE = "large"

model = whisper.load_model(MODEL_TYPE)

start = time.time()
file_name = "video"
extension = "mp4"
file_path = f"data/{file_name}.{extension}"
print(f"transcribing: {file_path}")

base_result = model.transcribe(file_path)

wc = len(base_result["text"].split(" "))

end = time.time()
total_seconds = end - start

print(f"total words: {wc}")
print(f"total seconds: {total_seconds}")
print(f"words/seconds: {wc}/{total_seconds}={wc/total_seconds}")

# save ['text']
with open(f'{file_name}-transcript-{MODEL_TYPE}.txt', 'w') as myfile:
    myfile.write(base_result['text'])

# save ['segments']
with open(f'{file_name}-segments-{MODEL_TYPE}.json', 'w') as myfile:
    myfile.write(json.dumps(base_result['segments']))

def get_segments_formatted(segments):
    def seconds_to_HH_MM_SS(duration_seconds):
      minutes, seconds = divmod(duration_seconds, 60)
      hours, minutes = divmod(minutes, 60)

      formatted_duration = "{:02d}:{:02d}:{:02d}".format(int(hours), int(minutes), int(seconds))
      return formatted_duration

    segment_transcript = []
    for segment in segments:
      start = seconds_to_HH_MM_SS(segment['start'])
      text = segment['text']
      print(f"{start} {text}")
      segment_transcript.append(f"{start} {text}")

    return "\n".join(segment_transcript)

# save ['segments'] but format the time
with open(f'{file_name}-sentence-transcript-{MODEL_TYPE}.txt', 'w') as myfile:

    myfile.write(get_segments_formatted(base_result['segments']))