From eb8f285b3952858654b0171c536dc6f78f3ff9bb Mon Sep 17 00:00:00 2001 From: Ziwen <67124639+zw123han@users.noreply.github.com> Date: Wed, 17 May 2023 23:19:40 -0400 Subject: [PATCH 1/2] Fix prefix bug when timestamps enabled --- whisper/decoding.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/whisper/decoding.py b/whisper/decoding.py index 457ee7c..7904d46 100644 --- a/whisper/decoding.py +++ b/whisper/decoding.py @@ -586,6 +586,8 @@ class DecodingTask: if self.sample_len is not None: max_prefix_len = self.n_ctx // 2 - self.sample_len prefix_tokens = prefix_tokens[-max_prefix_len:] + if not self.without_timestamps: + tokens = tokens + [self.tokenizer.timestamp_begin] tokens = tokens + prefix_tokens if prompt := self.options.prompt: From 5b16d67519db848a2bc2fccbffab2d785959d4c6 Mon Sep 17 00:00:00 2001 From: Ziwen <67124639+zw123han@users.noreply.github.com> Date: Wed, 17 May 2023 23:20:53 -0400 Subject: [PATCH 2/2] Update decoding.py --- whisper/decoding.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/whisper/decoding.py b/whisper/decoding.py index 7904d46..d5a4660 100644 --- a/whisper/decoding.py +++ b/whisper/decoding.py @@ -586,7 +586,7 @@ class DecodingTask: if self.sample_len is not None: max_prefix_len = self.n_ctx // 2 - self.sample_len prefix_tokens = prefix_tokens[-max_prefix_len:] - if not self.without_timestamps: + if not self.options.without_timestamps: tokens = tokens + [self.tokenizer.timestamp_begin] tokens = tokens + prefix_tokens