Skip to content

whisper: don't force a timestamp after 16 tokens — truncated every transcription (#698) #699

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Anjielon wants to merge 1 commit into
base: main
Choose a base branch
from
Open

whisper: don't force a timestamp after 16 tokens — truncated every transcription (#698) #699

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion src/common/whisper/modeling_whisper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,7 +182,11 @@ std::pair<std::string, std::string> Whisper::generate(whisper_task_type_t task,
os << token_str << std::flush;
}

int watching_dog = 16;
// Forcing a timestamp after only 16 content tokens (~4 s of speech) makes
// Whisper close the segment and emit EOT: every transcription was cut at
// ~15 tokens (see issue #698). 448 = decoder max, so the guard only fires
// on a genuinely runaway decode.
int watching_dog = 448;
for (int i = 0; i < 448 - 3; i++){
buffer<bf16> logits = this->whisper_engine->decode_audio(last_idx);
if (watching_dog == 0 && allow_force_time_stamp){
Expand Down