fx negative ths bug

This commit is contained in:
adamnsandle
2024-11-25 13:54:46 +00:00
parent cff5eb2980
commit eaf633ec9d

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