feat: ship Wave 5-6 multi-surface hygiene, optimization, and bug fixes

This commit is contained in:
Yun Chan 2026-11-14 11:51:41 +09:00
parent 01ef3293cf
commit 66b6b45ba9
399 changed files with 17828 additions and 11330 deletions

View file

@ -3,6 +3,7 @@
// 데스크톱은 전사를 `[MM:SS] [화자] 내용` 줄로 저장하고, Supabase `transcripts` 는 구간 행으로 저장한다
// (모바일은 구간이 있으면 구간을 우선 표시한다). 두 표현을 이 한 곳에서만 바꾼다.
import { formatDurationClockFromMs } from './format'
import { splitTextToFit } from './text-chunking'
export interface TranscriptLine {
@ -38,10 +39,7 @@ export function parseTranscriptLines(text: string | null | undefined): Transcrip
}
export function formatTranscriptTimestamp(ms: number): string {
const totalSeconds = Math.max(0, Math.floor(ms / 1000))
const minutes = Math.floor(totalSeconds / 60)
const seconds = totalSeconds % 60
return `${String(minutes).padStart(2, '0')}:${String(seconds).padStart(2, '0')}`
return formatDurationClockFromMs(ms, { padMinutes: true })
}
/** parseTranscriptLines 의 역변환(시각 표기가 없던 줄은 그대로). */
@ -74,11 +72,15 @@ export function buildTranscriptFromSegments(
sessionStartedAt: number,
): string {
const lines: TranscriptLine[] = []
const safeStartedAt = Number.isFinite(sessionStartedAt) ? sessionStartedAt : 0
for (const segment of segments) {
const text = segment.text.trim()
if (!text) continue
const timestampMs = Number.isFinite(segment.timestamp) && Number.isFinite(sessionStartedAt)
? Math.max(0, segment.timestamp - safeStartedAt)
: 0
lines.push({
timestampMs: Math.max(0, segment.timestamp - sessionStartedAt),
timestampMs,
timed: true,
speaker: null,
text,
@ -181,3 +183,67 @@ export function looksTruncatedRewrite(input: string, output: string): boolean {
}
return trimmed.length < input.trim().length * 0.5
}
const PYANNOTE_SPEAKER_PATTERN = /^SPEAKER_(\d+)$/iu
const TRAILING_SPEAKER_NUMBER_PATTERN = /(\d+)$/u
/**
* pyannote 원시 화자 ID(`SPEAKER_00`, `SPEAKER_01`)를 1 기반 표시 라벨(`화자 1`, `화자 2`)로 바꾼다.
* 이미 사용자 정의 이름이나 표시 라벨인 경우 공백만 정리해 그대로 돌려준다.
*/
export function formatPyannoteSpeakerLabel(rawSpeaker: string, prefix = '화자'): string {
const trimmed = rawSpeaker.trim()
const match = PYANNOTE_SPEAKER_PATTERN.exec(trimmed)
if (!match) return trimmed
const zeroBased = Number.parseInt(match[1], 10)
return `${prefix} ${zeroBased + 1}`
}
/**
* 화자 라벨(`화자 1`, `Speaker 2`, `SPEAKER_00`, `김민수`)을 팔레트 길이 내 결정적 색상 인덱스로 매핑한다.
*/
export function resolveSpeakerColorIndex(speaker: string, paletteSize = 5): number {
const size = Math.max(1, Math.floor(paletteSize))
const normalized = speaker.trim()
if (!normalized || size === 1) return 0
const pyannoteMatch = PYANNOTE_SPEAKER_PATTERN.exec(normalized)
if (pyannoteMatch) {
return Number.parseInt(pyannoteMatch[1], 10) % size
}
const trailingMatch = TRAILING_SPEAKER_NUMBER_PATTERN.exec(normalized)
if (trailingMatch) {
const rawDigits = trailingMatch[1]
const parsed = Number.parseInt(rawDigits, 10)
if (rawDigits.startsWith('0') || parsed === 0) {
return parsed % size
}
return (parsed - 1) % size
}
let hash = 2166136261
const lower = normalized.toLowerCase()
for (let i = 0; i < lower.length; i += 1) {
hash ^= lower.charCodeAt(i)
hash = Math.imul(hash, 16777619)
}
return (hash >>> 0) % size
}
/**
* 세그먼트 목록에서 화자 라벨을 첫 등장 순서대로 중복 없이 추출한다.
*/
export function extractDistinctSpeakers(
segments: readonly { readonly speaker?: string | null }[],
): string[] {
const seen = new Set<string>()
const speakers: string[] = []
for (const segment of segments) {
const label = segment.speaker?.trim()
if (!label || seen.has(label)) continue
seen.add(label)
speakers.push(label)
}
return speakers
}