Issue with Audio Segmentation When Video Contains Long Periods of Silence

[Shadows97]

There is an issue with the current audio segmentation approach in PySubtitle when processing videos that contain long periods of silence. The split_on_silence function from pydub may not handle these long silent segments effectively, leading to incomplete or inaccurate transcription and subtitle generation.

Steps to Reproduce:

1. Use a video file that contains long periods of silence (e.g., 5-10 seconds or more).
2. Run the audio_to_text function to convert the audio to text.
3. Observe the generated VTT file and note that the transcription may stop prematurely or miss segments of the video.

Expected Behavior:

The audio segmentation should handle long periods of silence more effectively, ensuring that the entire video is processed and transcribed accurately.

Actual Behavior: The transcription process may stop prematurely or miss segments of the video when long periods of silence are encountered.

Possible Solution:

• Adjust the parameters of the split_on_silence function to better handle long periods of silence.
• Implement a custom segmentation approach that can detect and handle long silent segments more effectively.
• Consider adding a fallback mechanism to ensure that the entire video is processed, even if long silent segments are present.

• Additional Context:

This issue affects the accuracy and completeness of the generated subtitles, especially for videos with significant silent segments. Improving the segmentation approach will enhance the overall reliability of PySubtitle.

[hcm444]

We can implement a custom segmentation method that detects and splits the audio based on a threshold duration of silence rather than relying solely on split_on_silence from pydub. So something like

def custom_split_on_silence(sound, min_silence_len=500, silence_thresh=-40)

[Shadows97]

possible