feat: Desktop版のElectron基盤と統合API仕様を追加

Backend/README.md

@@ -132,6 +132,13 @@ Cloud Run へのデプロイ手順は以下を参照してください。
     }
     ```
 
+- `POST /pipeline/transcribe-analyze`
+  - Desktop向け統合API。音声チャンクを受け取り、文字起こし（MVPでは `text_override` 指定可）と解析結果を1レスポンスで返す
+  - `is_final_chunk=false` の場合は解析をスキップし、partialテキストのみ返す
+  - `is_final_chunk=true` の場合は内容語ベクトル化・文ベクトル化・（必要時）辞書参照を返す
+  - 主要フォーム項目: `audio`, `session_id`, `chunk_seq`, `is_final_chunk`, `input_source`, `audio_format`, `include_dictionary`
+  - `audio_format` は `wav | pcm16 | webm_opus` をサポート
+
 - `POST /dictionary/lookup`
   - 用語の意味を日本語で1〜2文の概要として返す
   - このエンドポイントは現在DB参照未連携のため、常にGeminiで生成する
@@ -210,15 +217,18 @@ Backend/
     │   └── endpoints/
     │       ├── analysis.py
     │       ├── dictionary.py
-    │       └── hoge.py
+    │       ├── hoge.py
+    │       └── pipeline.py
     ├── services/
     │   ├── dictionary.py
+    │   ├── speech_pipeline.py
     │   ├── text_analysis.py
     │   └── hoge.py
     └── schemas/
         ├── analysis.py
         ├── dictionary.py
-        └── hoge.py
+        ├── hoge.py
+        └── pipeline.py
 ```
 
 ## 今後の実装予定

Backend/app/api/__init__.py

@@ -1,11 +1,12 @@
 import fastapi
-from app.api.endpoints import hoge, analysis, dictionary
+from app.api.endpoints import hoge, analysis, dictionary, pipeline
 
 router = fastapi.APIRouter()
 
 router.include_router(hoge.router, prefix="/hoge", tags=["hoge"])
 router.include_router(analysis.router, prefix="/analysis", tags=["analysis"])
 router.include_router(dictionary.router, prefix="/dictionary", tags=["dictionary"])
+router.include_router(pipeline.router, prefix="/pipeline", tags=["pipeline"])
 
 # 新しくエンドポイントを追加するときは、
 # 1. app/api/endpoints/new_endpoint.pyを作成する

Backend/app/api/endpoints/pipeline.py

@@ -0,0 +1,81 @@
+from __future__ import annotations
+
+import fastapi
+
+from app.schemas.pipeline import PipelineTranscribeAnalyzeResponse
+from app.services.speech_pipeline import transcribe_and_analyze_chunk
+
+router = fastapi.APIRouter()
+
+
+@router.post(
+    "/transcribe-analyze",
+    response_model=PipelineTranscribeAnalyzeResponse,
+    summary="音声チャンクを文字起こしし、解析結果をまとめて返す",
+    description=(
+        "Desktop向けの統合パイプラインAPIです。"
+        " 音声チャンクを受け取り、文字起こし（暫定）と解析結果を1レスポンスで返します。"
+    ),
+    responses={
+        200: {"description": "処理成功"},
+        400: {"description": "チャンク連番不正"},
+        413: {"description": "音声チャンクサイズ上限超過"},
+        415: {"description": "非対応フォーマット"},
+        422: {"description": "入力バリデーションエラー"},
+    },
+)
+async def transcribe_analyze(
+    audio: fastapi.UploadFile = fastapi.File(..., description="音声チャンク"),
+    session_id: str = fastapi.Form(..., min_length=1),
+    chunk_seq: int = fastapi.Form(..., ge=0),
+    is_final_chunk: bool = fastapi.Form(False),
+    input_source: str = fastapi.Form("microphone"),
+    audio_format: str = fastapi.Form("wav"),
+    sample_rate_hz: int = fastapi.Form(16000, ge=1),
+    channels: int = fastapi.Form(1, ge=1),
+    language_hint: str = fastapi.Form("ja-JP"),
+    include_dictionary: bool = fastapi.Form(False),
+    dictionary_top_k: int = fastapi.Form(5, ge=1, le=50),
+    deduplicate: bool = fastapi.Form(False),
+    min_length: int = fastapi.Form(1, ge=1, le=64),
+    normalize_sentence_vector: bool = fastapi.Form(True),
+    text_override: str | None = fastapi.Form(
+        default=None,
+        description="MVP向け暫定: STT未接続時に文字起こし結果として扱うテキスト",
+    ),
+) -> PipelineTranscribeAnalyzeResponse:
+    if input_source not in {"microphone", "system_audio"}:
+        raise fastapi.HTTPException(status_code=422, detail="input_source must be microphone or system_audio")
+
+    if audio_format not in {"wav", "pcm16", "webm_opus"}:
+        raise fastapi.HTTPException(status_code=415, detail="unsupported audio format")
+
+    normalized_session_id = session_id.strip()
+    if not normalized_session_id:
+        raise fastapi.HTTPException(status_code=422, detail="session_id must not be blank")
+
+    max_bytes = 5 * 1024 * 1024
+    audio_bytes = await audio.read(max_bytes + 1)
+    if not audio_bytes:
+        raise fastapi.HTTPException(status_code=422, detail="audio chunk is empty")
+    if len(audio_bytes) > max_bytes:
+        raise fastapi.HTTPException(status_code=413, detail="audio chunk too large")
+
+    result = await transcribe_and_analyze_chunk(
+        audio_bytes=audio_bytes,
+        session_id=normalized_session_id,
+        chunk_seq=chunk_seq,
+        is_final_chunk=is_final_chunk,
+        include_dictionary=include_dictionary,
+        dictionary_top_k=dictionary_top_k,
+        deduplicate=deduplicate,
+        min_length=min_length,
+        normalize_sentence_vector=normalize_sentence_vector,
+        text_override=text_override,
+    )
+
+    # 現時点のMVPではSTT実装に直結しないフィールドも受け付ける。
+    # 将来の実STT差し替え時にそのまま利用できるよう、ここではバリデーションのみ行う。
+    _ = (sample_rate_hz, channels, language_hint)
+
+    return PipelineTranscribeAnalyzeResponse(**result)

Backend/app/schemas/pipeline.py

@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+from pydantic import BaseModel, Field
+
+from app.schemas.analysis import (
+    ReferDictionaryEntry,
+    SentenceVectorizeResponse,
+    VectorizeResponse,
+)
+
+
+class PipelineTiming(BaseModel):
+    processed_ms: int = Field(ge=0, description="サーバー側処理時間（ミリ秒）")
+
+
+class PipelineTranscript(BaseModel):
+    partial_text: str = Field(default="", description="途中文字起こし")
+    final_text: str = Field(default="", description="確定文字起こし")
+    is_final: bool = Field(description="確定結果か")
+    confidence: float | None = Field(default=None, description="STT信頼度（未提供時はnull）")
+
+
+class PipelineAnalysis(BaseModel):
+    vectorize: VectorizeResponse | None = Field(
+        default=None,
+        description="内容語ベクトル化結果（非確定チャンクではnull）",
+    )
+    sentence_vectorize: SentenceVectorizeResponse | None = Field(
+        default=None,
+        description="文章ベクトル化結果（非確定チャンクではnull）",
+    )
+
+
+class PipelineDictionary(BaseModel):
+    enabled: bool = Field(description="辞書参照を実行したか")
+    entries: list[ReferDictionaryEntry] = Field(default_factory=list, description="辞書参照結果")
+
+
+class PipelineTranscribeAnalyzeResponse(BaseModel):
+    session_id: str = Field(description="セッション識別子")
+    chunk_seq: int = Field(ge=0, description="チャンク連番")
+    timing: PipelineTiming
+    transcript: PipelineTranscript
+    analysis: PipelineAnalysis
+    dictionary: PipelineDictionary