mirror of
https://github.com/ggml-org/whisper.cpp.git
synced 2026-07-30 14:40:55 -05:00
The current implementation in `whisper_wrap_segment()` uses `strlen()` to count bytes, not UTF-8 characters. When splitting segments at `max_len`, this can break multi-byte UTF-8 characters, resulting in invalid sequences displayed as `�` (U+FFFD replacement character).