feat(meeting): ground AI transcript correction in learned vocabulary
AI polishing of a recorded or pasted transcript was a generic rewrite: it cleaned fillers but left names and in-house terms mangled even when the app already knew them. Polishing now corrects dictionary pronunciations and typed identifiers deterministically before and after the model, and hands the model the learned term list plus the user's own expressions with explicit rules: rewrite a misheard name, product or in-house term to the learned spelling, and never swap an ordinary noun or ending on similarity alone. Dictation output is re-checked the same way after the model runs.
This commit is contained in:
parent
1f5372bfcc
commit
e2d17601a4
8 changed files with 120 additions and 23 deletions
|
|
@ -152,6 +152,19 @@ describe('LLM 교열 참고 블록', () => {
|
|||
expect(section).toContain('오늘 회의는 세 시에 시작합니다')
|
||||
})
|
||||
|
||||
it('자주 쓰는 용어와 교정 지시를 함께 넣는다', () => {
|
||||
const glossary = buildTranscriptGlossary({
|
||||
dictionary: [{ word: 'D3RO', pronunciation: '디에로' }],
|
||||
phrases: phrases(['하람로그에 정리해줘', 6], ['안녕하세요', 4], ['dkssudgktpdy', 3]),
|
||||
})
|
||||
|
||||
const section = buildGlossaryPromptSection(glossary)
|
||||
|
||||
expect(section).toContain('반드시 아래 표기로 고치세요')
|
||||
expect(section).toContain('[사용자가 자주 쓰는 용어')
|
||||
expect(section).toContain('하람로그')
|
||||
expect(section).not.toContain('dkssudgktpdy')
|
||||
})
|
||||
it('STT 힌트는 기존 문맥 뒤에 붙는다', () => {
|
||||
const glossary = buildTranscriptGlossary({ dictionary: [{ word: 'Whisper' }] })
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue