Merge pull request #581 from snakers4/adamnsandle

fx negative ths bug
This commit is contained in:
Dimitrii Voronin
2024-11-25 16:55:38 +03:00
committed by GitHub

View File

@@ -308,7 +308,7 @@ def get_speech_timestamps(audio: torch.Tensor,
current_speech = {}
if neg_threshold is None:
neg_threshold = threshold - 0.15
neg_threshold = max(threshold - 0.15, 0.01)
temp_end = 0 # to save potential segment end (and tolerate some silence)
prev_end = next_start = 0 # to save potential segment limits in case of maximum segment size reached