[Bug] Incorrect Timestamps in Subtitles for YouTube Video

[Martouta]

Description

I've encountered an issue with incorrect timestamps in the subtitles generated for a YouTube video using the speech_processor tool.

Steps to Reproduce

1. Prepare an input JSON file named xjEFo3a1AnI_input.json with the following content:

   [
     {
       "integration": "youtube",
       "id": "xjEFo3a1AnI",
       "language_code": "en-US",
       "resource_id": 7231,
       "recognizer": "google",
       "captions": "try"
     }
   ]   

Note: The YouTube video in question exists and has manual captions in English (US).

2. Run the speech_processor with the following command:

LOG_OUTPUT=standard MAX_THREADS=8 INPUT_FILE='xjEFo3a1AnI_input.json' SPEECH_ENV='development' SUBS_LOCATION='file' python3 .

3. After the process completes, inspect the generated subtitles in resources/subtitles/development.

$ cd resources/subtitles/development
$ grep -n "1038" 123145579446272-7231-11-11.18:03:26723463-subs.srt | cut -d: -f1 | xargs -I {} awk 'NR>={}-5 && NR<={}+5'  123145579446272-7231-11-11.18:03:26723463-subs.srt

Observed Behavior

The timestamps in the generated subtitles file 123145579446272-7231-11-11.18:03:26723463-subs.srt are incorrect. For instance, the following excerpt shows an issue with the timestamps:

1037
00:00:02,731 --> 00:00:02,736
Usually, for depression, I
want to see at least greater

1038
00:00:02,736 --> 00:00:02,739
than probably 0.8 minimal.

1039
00:00:02,739 --> 00:00:02,744

Expected Behavior

The timestamps in the subtitles should accurately reflect the timing of the spoken words in the video.

Additional Information

Environment: Development
Tool Version: v3.0.0
Python Version: Python 3.10.0
Operating System: macOS 14.1.1