Vocabulary cleaning strips a closing parenthesis and a trailing period from terms sent to Whisper #574

New issue

Open

opened 2026-09-09 06:43:46 +00:00 by claude-bot · 0 comments

claude-bot commented

2026-09-09 06:43:46 +00:00

Contributor

vocabulary_service.clean_term strips _EDGE_PUNCTUATION from both ends of a term, so the ASR receives Viquilonto (Viq (unbalanced) and Whispering Lantern Supply Co instead of the campaign's actual names. Seen in the 36-term list sent for the 2026-09-09 session, on prod and on the dev replay alike.

The intent (drop a stray comma or quote a GM typed) is right; the rule is too broad. Keep edge stripping for quotes, commas, semicolons and dashes, but leave a closing bracket that has its opener inside the term, and leave a trailing period when the term ends in an abbreviation (Co., Mt.). Two tests: the two names above survive intact; "Aramore," still becomes Aramore.

Low severity on its own; filed because an unbalanced parenthesis in an initial_prompt is one more thing for Whisper to echo (#569).

`vocabulary_service.clean_term` strips `_EDGE_PUNCTUATION` from both ends of a term, so the ASR receives `Viquilonto (Viq` (unbalanced) and `Whispering Lantern Supply Co` instead of the campaign's actual names. Seen in the 36-term list sent for the 2026-09-09 session, on prod and on the dev replay alike. The intent (drop a stray comma or quote a GM typed) is right; the rule is too broad. Keep edge stripping for quotes, commas, semicolons and dashes, but leave a closing bracket that has its opener inside the term, and leave a trailing period when the term ends in an abbreviation (`Co.`, `Mt.`). Two tests: the two names above survive intact; `"Aramore,"` still becomes `Aramore`. Low severity on its own; filed because an unbalanced parenthesis in an `initial_prompt` is one more thing for Whisper to echo (#569).

claude-bot referenced this issue

2026-09-09 08:25:02 +00:00

Send the campaign vocabulary as Whisper hotwords instead of an initial_prompt, behind a switch, and measure the echo rate #577

claude-bot referenced this issue

2026-09-09 10:49:21 +00:00

Snap misheard character names to GM-defined "heard as" aliases after transcription #579