Best way to cut a long sentence in half?

[Francoyy]

Super excited about using this new tool, there are many features that can help my subtitle generation for both English and Chinese videos!
One question, in my testing, there was about 20% of time sequences that were just very long compared to most other sequences in the SRT file.
I'm trying to address those very long sentences.
Ideally, I would like those very long sentences to be cut in half, roughly from the middle, where it makes the most sense to cut them.
I tried a different number of parameters, such as:
--segment_level true --word_level false --max_chars 56
--segment_level true --word_level false --max_chars 56 --regroup sl
--segment_level true --word_level false --max_chars 56 --regroup sp

I didn't have so much luck. One example is as such:

250
00:09:48,560 --> 00:09:53,040
Yes, so I think the key thing is really to maximize the

251
00:09:53,080 --> 00:09:53,440
savings

I hope that the sentence could have been chopped more around the middle. Is there any option that would do something like that?

One possible option in mind, is that I would add --word_level true, and since I have the timestamps for every word, I think I could be able to chop a long sentence in half while preserving correct timestamps. But that's not perfect, because cutting in half won't guarantee it cuts in the best place.

Thanks!

[snakers4]

Hi, this looks like an ASR question, not a VAD question.

The VAD is "stupid", i.e. it only says whether there is speech or not.

We provide the following params:

silero-vad/utils_vad.py

Lines 164 to 225 in 563106e

	def get_speech_timestamps(audio: torch.Tensor,
	model,
	threshold: float = 0.5,
	sampling_rate: int = 16000,
	min_speech_duration_ms: int = 250,
	max_speech_duration_s: float = float('inf'),
	min_silence_duration_ms: int = 100,
	window_size_samples: int = 512,
	speech_pad_ms: int = 30,
	return_seconds: bool = False,
	visualize_probs: bool = False,
	progress_tracking_callback: Callable[[float], None] = None):
	"""
	This method is used for splitting long audios into speech chunks using silero VAD
	Parameters
	----------
	audio: torch.Tensor, one dimensional
	One dimensional float torch.Tensor, other types are casted to torch if possible
	model: preloaded .jit silero VAD model
	threshold: float (default - 0.5)
	Speech threshold. Silero VAD outputs speech probabilities for each audio chunk, probabilities ABOVE this value are considered as SPEECH.
	It is better to tune this parameter for each dataset separately, but "lazy" 0.5 is pretty good for most datasets.
	sampling_rate: int (default - 16000)
	Currently silero VAD models support 8000 and 16000 sample rates
	min_speech_duration_ms: int (default - 250 milliseconds)
	Final speech chunks shorter min_speech_duration_ms are thrown out
	max_speech_duration_s: int (default - inf)
	Maximum duration of speech chunks in seconds
	Chunks longer than max_speech_duration_s will be split at the timestamp of the last silence that lasts more than 100ms (if any), to prevent agressive cutting.
	Otherwise, they will be split aggressively just before max_speech_duration_s.
	min_silence_duration_ms: int (default - 100 milliseconds)
	In the end of each speech chunk wait for min_silence_duration_ms before separating it
	window_size_samples: int (default - 1536 samples)
	Audio chunks of window_size_samples size are fed to the silero VAD model.
	WARNING! Silero VAD models were trained using 512, 1024, 1536 samples for 16000 sample rate and 256, 512, 768 samples for 8000 sample rate.
	Values other than these may affect model perfomance!!
	speech_pad_ms: int (default - 30 milliseconds)
	Final speech chunks are padded by speech_pad_ms each side
	return_seconds: bool (default - False)
	whether return timestamps in seconds (default - samples)
	visualize_probs: bool (default - False)
	whether draw prob hist or not
	progress_tracking_callback: Callable[[float], None] (default - None)
	callback function taking progress in percents as an argument
	Returns
	----------
	speeches: list of dicts
	list containing ends and beginnings of speech chunks (samples or seconds based on return_seconds)