Mirroristas/whisper
2
0
Fork 0
mirror of https://github.com/openai/whisper.git synced 2025-11-23 22:15:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

word timing tweaks (#1559)

Browse Source 
* word timing tweaks

* comment on eot

* clearer comments
This commit is contained in:
taylorchu 2023-08-07 14:48:56 -07:00 committed by GitHub
parent b91c907694
commit e8622f9afc
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 7 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
whisper/timing.py
View File

@ -214,6 +214,13 @@ def find_alignment(
text_indices, time_indices = dtw(-matrix) text_indices, time_indices = dtw(-matrix)
words, word_tokens = tokenizer.split_to_word_tokens(text_tokens + [tokenizer.eot]) words, word_tokens = tokenizer.split_to_word_tokens(text_tokens + [tokenizer.eot])
if len(word_tokens) <= 1:
# return on eot only
# >>> np.pad([], (1, 0))
# array([0.])
# This results in crashes when we lookup jump_times with float, like
# IndexError: arrays used as indices must be of integer (or boolean) type
return []
word_boundaries = np.pad(np.cumsum([len(t) for t in word_tokens[:-1]]), (1, 0)) word_boundaries = np.pad(np.cumsum([len(t) for t in word_tokens[:-1]]), (1, 0))
jumps = np.pad(np.diff(text_indices), (1, 0), constant_values=1).astype(bool) jumps = np.pad(np.diff(text_indices), (1, 0), constant_values=1).astype(bool)
@ -297,8 +304,6 @@ def add_word_timestamps(
# hack: truncate long words at sentence boundaries. # hack: truncate long words at sentence boundaries.
# a better segmentation algorithm based on VAD should be able to replace this. # a better segmentation algorithm based on VAD should be able to replace this.
if len(word_durations) > 0: if len(word_durations) > 0:
median_duration = np.median(word_durations)
max_duration = median_duration * 2
sentence_end_marks = ".。!！?？" sentence_end_marks = ".。!！?？"
# ensure words at sentence boundaries are not longer than twice the median word duration. # ensure words at sentence boundaries are not longer than twice the median word duration.
for i in range(1, len(alignment)): for i in range(1, len(alignment)):