The SRTConverter is a Python class designed to manipulate segments from a WhisperX transcription into SRT (SubRip Text) format. This guide provides examples on how to use this class effectively.
To initialize the SRTConverter, you need a list of segment dictionaries. Each segment dictionary should contain start, end, text, and a list of words with their respective start, end times, and scores.
from SRTConverter import SRTConverter
segments = [
{
"start": 0.27,
"end": 1.632,
"text": "Hello world.",
"words": [
{"word": "Hello", "start": 0.27, "end": 0.61, "score": 0.862},
{"word": "world.", "start": 0.69, "end": 1.091, "score": 0.779}
],
},
# ... more segments ...
]
converter = SRTConverter(segments)To generate an SRT file with specific words highlighted:
srt_highlighted = converter.to_srt_highlight_word(color="red")
print(srt_highlighted)To create an SRT file where each word is a separate entry:
srt_single_words = converter.to_srt_single_words()
print(srt_single_words)To generate an SRT file with plain text segments:
srt_plain_text = converter.to_srt_plain_text()
print(srt_plain_text)Optionally, you can specify the number of words per segment:
srt_plain_text_custom = converter.to_srt_plain_text(words_per_segment=3)
print(srt_plain_text_custom)If you need to reorganize the word segments based on a specific number of words per segment:
new_segments = converter.adjust_word_per_segment(words_per_segment=4)To create new segments from a list of word dictionaries:
word_segments = [
# ... list of word dictionaries ...
]
new_segments = SRTConverter.create_segments_from_words(word_segments, words_per_segment=5)To write the generated SRT string to a file:
SRTConverter.write_to_file("output.srt", srt_highlighted)To check if an SRT string or file is valid:
is_valid_string = SRTConverter.is_valid_srt_string(srt_string)
is_valid_file = SRTConverter.is_valid_srt_file("output.srt")The initialize_with_normalized_timestamps static method of the SRTConverter class initializes the converter with multiple arrays of segments while normalizing their timestamps. This method is particularly useful in scenarios where a long audio or video file has been transcribed in multiple parts (perhaps in parallel), and there's a need to combine these segmented transcriptions into one continuous SRT file.
By normalizing timestamps, this method ensures that each segment follows sequentially from the end of the previous segment, regardless of the starting timestamp of each individual transcription.
segment_arrays(list of lists): A list containing arrays of segment dictionaries. Each array represents a separate part of the transcription, and each segment dictionary within the array should contain keys forstart,end,text, and optionallywords.
SRTConverter: An initialized instance ofSRTConverterwith segments that have normalized timestamps.
NOTE NEED TO FIX. IMPLEMENTATION IS WRONG.
I assumed the last transcription time from A would be the end time of the audio, but that is not true. I need to check the actual audio lengths.
# Assuming you have multiple arrays of segments from parallel transcription processes
segment_array_1 = [
# ... segments from the first part of the transcription ...
]
segment_array_2 = [
# ... segments from the second part of the transcription ...
]
# Initialize the SRTConverter with normalized timestamps
converter = SRTConverter.initialize_with_normalized_timestamps([segment_array_1, segment_array_2])
# Now the converter.segments will have a continuous timeline combining segment_array_1 and segment_array_2https://github.com/m-bain/whisperX?tab=readme-ov-file#python-usage--
# 1. Transcribe with original whisper (batched)
model = whisperx.load_model("large-v2", device, compute_type=compute_type)
audio = whisperx.load_audio(audio_file)
result = model.transcribe(audio, batch_size=batch_size)
print(result["segments"]) # before alignment
# delete model if low on GPU resources
# import gc; gc.collect(); torch.cuda.empty_cache(); del model
# 2. Align whisper output
model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)
result = whisperx.align(result["segments"], model_a, metadata, audio, device, return_char_alignments=False)
print(result["segments"]) # after alignment
# 3. Initialize SRTConverter and generate SRT to highlight words
converter = SRTConverter(result["segments"])
highlighted_srt = converter.to_srt_highlight_word(color="red")
SRTConverter.write_to_file("output.srt", highlighted_srt)