feat(V2-1a): Monorepo 구조 전환 — apps/desktop으로 V1 이동

- npm workspaces 루트 (apps/*, packages/*) 세팅 - V1 전체를 apps/desktop/으로 git mv (src, resources, tests, sidecar, scripts, electron.vite.config.ts, electron-builder.yml, vitest.config.ts, tsconfig.node.json, tsconfig.web.json) - apps/desktop/package.json 신규 (name=@d3ro/desktop) - productName: 'd3ro-voice' 명시 — app.getName()을 고정하여 userData 경로 %APPDATA%\d3ro-voice\ 그대로 유지 (기존 DB/설정 연속성 보장) - 루트 package.json을 workspace 루트로 재구성, 공통 devDep만 유지 (typescript, eslint, prettier) - turbo.json, tsconfig.base.json 추가 (Turborepo 자체 설치는 별도 sub-phase) - memory/project_status.md 생성 (규칙 13) 검증: - npm run typecheck 통과 - npm run build 통과 (electron-vite main+preload+renderer) - npm run dev 실제 실행 → DB/핫키/Ollama 자동 실행 모두 정상
2026-04-08 14:04:41 +09:00 · 2026-04-08 14:04:41 +09:00 · 45a580878a
commit 45a580878a
parent 3a160b9032
178 changed files with 214 additions and 0 deletions
--- a/apps/desktop/src/main/services/CaptionService.ts
+++ b/apps/desktop/src/main/services/CaptionService.ts
@ -0,0 +1,557 @@
+// src/main/services/CaptionService.ts
+// Phase 10.1: Live Caption — 실시간 자막 서비스
+// 3초 청크 기반 스트리밍 전사. 싱글톤 + EventEmitter 패턴.
+
+import { EventEmitter } from 'events'
+import { nanoid } from 'nanoid'
+import { getLogger } from './LoggerService'
+import { getAudioCaptureService, calculateRMS } from './AudioCaptureService'
+import { getSoundEffectService } from './SoundEffectService'
+import { getLocalSTTService } from './LocalSTTService'
+import { getHistoryService } from './HistoryService'
+import { configGet } from './ConfigService'
+import {
+  showCaptionOverlay,
+  hideCaptionOverlay,
+  sendToCaptionOverlay,
+} from '../windows/WindowManager'
+import { IPC_CHANNELS } from '@shared/ipc-channels'
+import { D3ROError, ErrorCode } from '@shared/errors'
+import type {
+  CaptionState,
+  CaptionSegment,
+  CaptionConfig,
+  CaptionSessionSummary,
+} from '@shared/types'
+import { getMainWindow } from '../windows/WindowManager'
+
+const logger = getLogger('CaptionService')
+
+/** 청크 수집 간격 (ms) — 6초로 충분한 컨텍스트 확보 */
+const CHUNK_INTERVAL_MS = 6000
+
+/** RMS 무음 임계값 — 이하면 무음으로 판정 (SoX 캡처 레벨이 낮으므로 0.003 사용) */
+const SILENCE_RMS_THRESHOLD = 0.003
+
+/** 유성음 프레임 비율 — 이 비율 미만이면 청크 스킵 (환각 방지) */
+const VOICED_FRAME_RATIO = 0.03
+
+/** initialPrompt 컨텍스트 윈도우 (자) */
+const CONTEXT_WINDOW_SIZE = 300
+
+/** 기본 자막 설정 */
+const DEFAULT_CONFIG: CaptionConfig = {
+  fontSize: 18,
+  opacity: 0.85,
+  maxLines: 3,
+  autoClearMs: 5000,
+  audioSource: 'mic',
+}
+
+interface CaptionServiceEvents {
+  'state-changed': (state: CaptionState) => void
+  'segment': (segment: CaptionSegment) => void
+  'delta': (data: { text: string; isFinal: boolean }) => void
+  'session-saved': (summary: CaptionSessionSummary) => void
+  'error': (error: D3ROError) => void
+}
+
+class CaptionService extends EventEmitter {
+  private _state: CaptionState = 'inactive'
+  private _config: CaptionConfig = { ...DEFAULT_CONFIG }
+  private _sessionId: string | null = null
+  private _sessionStartedAt: number | null = null
+  private _segments: CaptionSegment[] = []
+  private _audioBuffers: Buffer[] = []
+  private _chunkTimer: ReturnType<typeof setInterval> | null = null
+  private _isProcessingChunk = false
+  private _previousContext = ''
+  private _disposed = false
+  /** 현재 청크 내 유성음 프레임 수 */
+  private _voicedFrameCount = 0
+  /** 현재 청크 내 총 프레임 수 */
+  private _totalFrameCount = 0
+
+  private _audioDataHandler: ((payload: { buffer: Buffer; timestamp: number }) => void) | null = null
+  /** 시스템 오디오용 별도 버퍼 (audioSource='system' 또는 'both') */
+  private _systemAudioBuffers: Buffer[] = []
+  private _systemVoicedFrameCount = 0
+  private _systemTotalFrameCount = 0
+
+  // ── 공개 접근자 ──
+
+  getState(): CaptionState {
+    return this._state
+  }
+
+  getConfig(): CaptionConfig {
+    return { ...this._config }
+  }
+
+  setConfig(partial: Partial<CaptionConfig>): void {
+    this._config = { ...this._config, ...partial }
+    logger.debug(`Caption config updated: ${JSON.stringify(this._config)}`)
+
+    // 오버레이에 설정 변경 알림
+    sendToCaptionOverlay('caption:config', this._config)
+  }
+
+  // ── 시작 ──
+
+  async start(): Promise<void> {
+    if (this._disposed) {
+      throw new D3ROError(
+        ErrorCode.CaptionStartFailed,
+        'CaptionService가 이미 dispose되었습니다',
+      )
+    }
+
+    if (this._state !== 'inactive') {
+      throw new D3ROError(
+        ErrorCode.CaptionAlreadyActive,
+        `캡션이 이미 활성 상태입니다: ${this._state}`,
+      )
+    }
+
+    this._setState('starting')
+
+    // 오버레이를 즉시 표시 (로딩 상태)
+    showCaptionOverlay()
+    sendToCaptionOverlay('caption:config', this._config)
+    sendToCaptionOverlay(IPC_CHANNELS.CAPTION.STATE_CHANGED, { state: 'starting' })
+
+    try {
+      // STT 초기화 (모델 로딩 — 시간 소요)
+      const sttService = getLocalSTTService()
+      const modelId = configGet('sttModelId') as string | undefined
+      await sttService.initialize(modelId)
+
+      // 세션 초기화
+      this._sessionId = nanoid()
+      this._sessionStartedAt = Date.now()
+      this._segments = []
+      this._audioBuffers = []
+      // 초기 컨텍스트 힌트 — Whisper 첫 청크 환각 방지
+      const lang = configGet('sttLanguage') as string | undefined
+      this._previousContext = lang === 'ko' ? '다음은 한국어 대화입니다.' : ''
+      this._isProcessingChunk = false
+      this._voicedFrameCount = 0
+      this._totalFrameCount = 0
+
+      // ConfigService에서 저장된 오디오 소스 읽기
+      const savedSource = configGet('captionAudioSource' as keyof import('@shared/types').AppConfig) as unknown as string
+      if (savedSource && (savedSource === 'mic' || savedSource === 'system' || savedSource === 'both')) {
+        this._config.audioSource = savedSource as 'mic' | 'system' | 'both'
+      }
+      const { audioSource } = this._config
+      logger.info(`Caption audioSource: ${audioSource}`)
+
+      // 마이크 캡처 (mic 또는 both)
+      if (audioSource === 'mic' || audioSource === 'both') {
+        const audioCaptureService = getAudioCaptureService()
+        this._audioDataHandler = (payload) => {
+          this._onAudioData(payload.buffer)
+        }
+        audioCaptureService.on('audio-data', this._audioDataHandler)
+        await audioCaptureService.start()
+      }
+
+      // 시스템 오디오 캡처 요청 (system 또는 both) — 렌더러에 시작 요청
+      if (audioSource === 'system' || audioSource === 'both') {
+        this._sendToMainWindow(IPC_CHANNELS.CAPTION.START_SYSTEM_AUDIO, {})
+        this._systemAudioBuffers = []
+        this._systemVoicedFrameCount = 0
+        this._systemTotalFrameCount = 0
+      }
+
+      // 청크 타이머 시작
+      this._chunkTimer = setInterval(() => {
+        const promises: Promise<void>[] = []
+
+        // 마이크 청크 처리
+        if (audioSource === 'mic' || audioSource === 'both') {
+          promises.push(
+            this._processChunk().catch((err: unknown) => {
+              logger.error(`마이크 청크 처리 실패: ${err instanceof Error ? err.message : String(err)}`)
+            })
+          )
+        }
+
+        // 시스템 오디오 청크 처리
+        if (audioSource === 'system' || audioSource === 'both') {
+          promises.push(
+            this._processSystemChunk().catch((err: unknown) => {
+              logger.error(`시스템 오디오 청크 처리 실패: ${err instanceof Error ? err.message : String(err)}`)
+            })
+          )
+        }
+
+        Promise.all(promises).catch(() => { /* 개별 에러는 이미 로깅됨 */ })
+      }, CHUNK_INTERVAL_MS)
+
+      this._setState('active')
+      getSoundEffectService().play('recording-start')
+      logger.info(`Live Caption 시작: sessionId=${this._sessionId}`)
+    } catch (err) {
+      this._setState('inactive')
+      const d3roErr =
+        err instanceof D3ROError
+          ? err
+          : new D3ROError(
+              ErrorCode.CaptionStartFailed,
+              `캡션 시작 실패: ${err instanceof Error ? err.message : String(err)}`,
+            )
+      this.emit('error', d3roErr)
+      throw d3roErr
+    }
+  }
+
+  // ── 중지 ──
+
+  async stop(): Promise<void> {
+    if (this._state !== 'active' && this._state !== 'starting') {
+      logger.warn(`캡션 중지 불가: 현재 상태=${this._state}`)
+      return
+    }
+
+    this._setState('stopping')
+
+    // 타이머 정리
+    if (this._chunkTimer) {
+      clearInterval(this._chunkTimer)
+      this._chunkTimer = null
+    }
+
+    // 잔여 오디오 처리
+    try {
+      await this._processChunk()
+    } catch (err) {
+      logger.warn(
+        `잔여 오디오 처리 실패: ${err instanceof Error ? err.message : String(err)}`,
+      )
+    }
+
+    // 마이크 캡처 정리
+    if (this._audioDataHandler) {
+      const audioCaptureService = getAudioCaptureService()
+      audioCaptureService.off('audio-data', this._audioDataHandler)
+      this._audioDataHandler = null
+      await audioCaptureService.stop()
+    }
+
+    // 시스템 오디오 캡처 중지 요청
+    this._sendToMainWindow(IPC_CHANNELS.CAPTION.STOP_SYSTEM_AUDIO, {})
+    this._systemAudioBuffers = []
+
+    // 오버레이 숨김
+    hideCaptionOverlay()
+
+    // 세션을 히스토리에 저장
+    const summary = this._saveSession()
+
+    this._sessionId = null
+    this._sessionStartedAt = null
+    this._audioBuffers = []
+    this._previousContext = ''
+
+    this._setState('inactive')
+    getSoundEffectService().play('recording-stop')
+    logger.info('Live Caption 중지')
+
+    if (summary) {
+      this.emit('session-saved', summary)
+      this._sendToMainWindow(IPC_CHANNELS.CAPTION.SESSION_SAVED, summary)
+    }
+  }
+
+  // ── dispose ──
+
+  async dispose(): Promise<void> {
+    if (this._disposed) return
+    this._disposed = true
+
+    if (this._state === 'active' || this._state === 'starting') {
+      await this.stop()
+    }
+
+    this.removeAllListeners()
+    logger.info('CaptionService disposed')
+  }
+
+  // ── 내부: 오디오 데이터 수집 ──
+
+  private _onAudioData(buffer: Buffer): void {
+    if (this._state !== 'active') return
+    this._audioBuffers.push(buffer)
+
+    // RMS 기반 유성음 감지 (환각 방지)
+    this._totalFrameCount++
+    const rms = calculateRMS(buffer)
+    if (rms >= SILENCE_RMS_THRESHOLD) {
+      this._voicedFrameCount++
+    }
+  }
+
+  /**
+   * 시스템 오디오 PCM16 데이터를 수신한다 (렌더러 → IPC → 여기).
+   * 렌더러의 systemAudioCapture.ts가 이미 16kHz mono PCM16으로 변환하여 전송.
+   */
+  onSystemAudioData(buffer: Buffer): void {
+    if (this._state !== 'active') return
+    this._systemAudioBuffers.push(buffer)
+
+    this._systemTotalFrameCount++
+    const rms = calculateRMS(buffer)
+    if (rms >= SILENCE_RMS_THRESHOLD) {
+      this._systemVoicedFrameCount++
+    }
+  }
+
+  // ── 내부: 3초 청크 처리 ──
+
+  private async _processChunk(): Promise<void> {
+    if (this._audioBuffers.length === 0) return
+    if (this._isProcessingChunk) return
+
+    this._isProcessingChunk = true
+
+    // VAD 카운터 리셋 (로깅용)
+    const voicedRatio = this._totalFrameCount > 0
+      ? this._voicedFrameCount / this._totalFrameCount
+      : 0
+    this._voicedFrameCount = 0
+    this._totalFrameCount = 0
+    logger.debug(`청크 처리: 유성음 비율 ${(voicedRatio * 100).toFixed(1)}%`)
+
+    // VAD 필터링은 faster-whisper 사이드카에서 처리 (vadFilter: true)
+    // SoX의 마이크 캡처 레벨이 매우 낮아 로컬 RMS 게이트는 신뢰 불가
+
+    try {
+      const merged = Buffer.concat(this._audioBuffers)
+      this._audioBuffers = []
+
+      // 최소 오디오 크기 확인 (100ms 분량 이상)
+      const minBytes = 16000 * 2 * 0.1 // 100ms @ 16kHz 16bit mono
+      if (merged.length < minBytes) {
+        return
+      }
+
+      const sttService = getLocalSTTService()
+      const language = configGet('sttLanguage') as string | undefined
+
+      const result = await sttService.transcribe(merged, {
+        language: language ?? 'auto',
+        initialPrompt: this._previousContext,
+        vadFilter: true,
+      })
+
+      if (!result.text || result.text.trim().length === 0) {
+        return
+      }
+
+      const segment: CaptionSegment = {
+        id: nanoid(),
+        text: result.text.trim(),
+        timestamp: Date.now(),
+        isFinal: true,
+      }
+
+      this._segments.push(segment)
+
+      // 다음 청크를 위한 컨텍스트 업데이트 (연속성 유지)
+      this._previousContext = result.text.trim().slice(-CONTEXT_WINDOW_SIZE)
+
+      // 이벤트 emit
+      this.emit('segment', segment)
+
+      // 오버레이 윈도우에 전송
+      sendToCaptionOverlay(IPC_CHANNELS.CAPTION.SEGMENT, segment)
+
+      // 메인 윈도우에 전송 (Dashboard 등에서 사용)
+      this._sendToMainWindow(IPC_CHANNELS.CAPTION.SEGMENT, segment)
+
+      logger.debug(`자막 세그먼트: "${segment.text.substring(0, 50)}"`)
+    } catch (err) {
+      const d3roErr =
+        err instanceof D3ROError
+          ? err
+          : new D3ROError(
+              ErrorCode.CaptionSTTFailed,
+              `자막 전사 실패: ${err instanceof Error ? err.message : String(err)}`,
+            )
+      logger.error(`자막 전사 에러: ${d3roErr.message}`)
+      this.emit('error', d3roErr)
+    } finally {
+      this._isProcessingChunk = false
+    }
+  }
+
+  // ── 내부: 시스템 오디오 청크 처리 ──
+
+  private async _processSystemChunk(): Promise<void> {
+    if (this._systemAudioBuffers.length === 0) return
+
+    // VAD 게이트
+    const voicedRatio = this._systemTotalFrameCount > 0
+      ? this._systemVoicedFrameCount / this._systemTotalFrameCount
+      : 0
+    this._systemVoicedFrameCount = 0
+    this._systemTotalFrameCount = 0
+
+    if (voicedRatio < VOICED_FRAME_RATIO) {
+      this._systemAudioBuffers = []
+      logger.debug(`시스템 오디오 청크 스킵: 유성음 비율 ${(voicedRatio * 100).toFixed(1)}%`)
+      return
+    }
+
+    const merged = Buffer.concat(this._systemAudioBuffers)
+    this._systemAudioBuffers = []
+
+    const minBytes = 16000 * 2 * 0.1
+    if (merged.length < minBytes) return
+
+    try {
+      const sttService = getLocalSTTService()
+      const language = configGet('sttLanguage') as string | undefined
+
+      const result = await sttService.transcribe(merged, {
+        language: language ?? 'auto',
+        initialPrompt: this._previousContext,
+        vadFilter: true,
+      })
+
+      if (!result.text || result.text.trim().length === 0) return
+
+      const segment: CaptionSegment = {
+        id: nanoid(),
+        text: result.text.trim(),
+        timestamp: Date.now(),
+        isFinal: true,
+      }
+
+      this._segments.push(segment)
+      this._previousContext = result.text.trim().slice(-CONTEXT_WINDOW_SIZE)
+
+      this.emit('segment', segment)
+      sendToCaptionOverlay(IPC_CHANNELS.CAPTION.SEGMENT, segment)
+      this._sendToMainWindow(IPC_CHANNELS.CAPTION.SEGMENT, segment)
+
+      logger.debug(`시스템 자막: "${segment.text.substring(0, 50)}"`)
+    } catch (err) {
+      logger.error(`시스템 오디오 전사 실패: ${err instanceof Error ? err.message : String(err)}`)
+    }
+  }
+
+  // ── 내부: 세션 저장 ──
+
+  private _saveSession(): CaptionSessionSummary | null {
+    if (!this._sessionId || !this._sessionStartedAt || this._segments.length === 0) {
+      return null
+    }
+
+    const endedAt = Date.now()
+    const totalDurationMs = endedAt - this._sessionStartedAt
+    const fullText = this._segments.map((s) => s.text).join(' ')
+    const wordCount = fullText.split(/\s+/).filter((w) => w.length > 0).length
+
+    // 히스토리에 저장 (mode: 'caption')
+    try {
+      const historyService = getHistoryService()
+      historyService.create({
+        originalText: fullText,
+        polishedText: null,
+        focusedApp: null,
+        focusedAppName: null,
+        focusedAppWindowTitle: null,
+        mode: 'caption',
+        status: 'completed',
+        errorCode: null,
+        audioLocalPath: null,
+        duration: totalDurationMs / 1000,
+        detectedLanguage: null,
+        micDevice: null,
+        wordCount,
+        sttModel: configGet('sttModelId') as string | null,
+        llmModel: null,
+        sttLatencyMs: null,
+        llmLatencyMs: null,
+        appVersion: '1.0.0',
+      })
+      logger.info(
+        `캡션 세션 저장: ${this._segments.length}개 세그먼트, ${wordCount}단어, ${Math.round(totalDurationMs / 1000)}초`,
+      )
+    } catch (err) {
+      logger.error(
+        `캡션 세션 저장 실패: ${err instanceof Error ? err.message : String(err)}`,
+      )
+    }
+
+    const summary: CaptionSessionSummary = {
+      sessionId: this._sessionId,
+      segments: [...this._segments],
+      startedAt: this._sessionStartedAt,
+      endedAt,
+      totalDurationMs,
+    }
+
+    this._segments = []
+    return summary
+  }
+
+  // ── 내부: 상태 전이 ──
+
+  private _setState(newState: CaptionState): void {
+    if (this._state === newState) return
+    const prev = this._state
+    this._state = newState
+    logger.debug(`CaptionState: ${prev} -> ${newState}`)
+    this.emit('state-changed', newState)
+    this._sendToMainWindow(IPC_CHANNELS.CAPTION.STATE_CHANGED, { state: newState })
+    sendToCaptionOverlay(IPC_CHANNELS.CAPTION.STATE_CHANGED, { state: newState })
+  }
+
+  // ── 내부: 메인 윈도우에 IPC 전송 ──
+
+  private _sendToMainWindow(channel: string, data: unknown): void {
+    const win = getMainWindow()
+    if (win && !win.isDestroyed()) {
+      win.webContents.send(channel, data)
+    }
+  }
+
+  // ── 타입 안전한 이벤트 메서드 오버라이드 ──
+
+  override emit<K extends keyof CaptionServiceEvents>(
+    event: K,
+    ...args: Parameters<CaptionServiceEvents[K]>
+  ): boolean {
+    return super.emit(event, ...args)
+  }
+
+  override on<K extends keyof CaptionServiceEvents>(
+    event: K,
+    listener: CaptionServiceEvents[K],
+  ): this {
+    return super.on(event, listener)
+  }
+
+  override off<K extends keyof CaptionServiceEvents>(
+    event: K,
+    listener: CaptionServiceEvents[K],
+  ): this {
+    return super.off(event, listener)
+  }
+}
+
+// ── 싱글톤 ──
+
+let instance: CaptionService | null = null
+
+export function getCaptionService(): CaptionService {
+  if (!instance) {
+    instance = new CaptionService()
+  }
+  return instance
+}
+
+export { CaptionService }