feat(stage11): audio integration — STT for interviews, podcasts, press conferences with timestamped claims

src/nsct/storage/models.py

@@ -538,7 +538,7 @@ class VisionEvidenceModel(Base):
 
     extracted_text = Column(Text, nullable=False)
     image_data_url = Column(Text, nullable=True)
-    entities = Column(JSON, nullable=False, default=dict)
+    extracted_entities = Column(JSON, nullable=False, default=dict)
 
     confidence = Column(Float, nullable=False, default=0.5)
     confidence_label = Column(String(16), nullable=False, default="medium")
@@ -601,4 +601,157 @@ VisionEvidenceModel.entities = relationship(
     back_populates="evidence",
     cascade="all, delete-orphan",
     foreign_keys="VisionEntityModel.evidence_id",
-)
+)
+
+
+# ---------------------------------------------------------------------------
+# Stage 11 — Audio Evidence Extraction (Speech-to-Text)
+# ---------------------------------------------------------------------------
+
+
+class AudioSegmentType(str, enum.Enum):
+    """Klassifikation der Audio-Quelle (Stage 11)."""
+
+    INTERVIEW = "interview"
+    PODCAST = "podcast"
+    PRESSEKONFERENZ = "pressekonferenz"
+    REDEN = "reden"
+    SONSTIGE = "sonstige"
+
+
+class AudioTranscriptModel(Base):
+    """Transkript eines Audio-Eintrags (Stage 11).
+
+    Felder:
+        uuid: Primärschlüssel (UUID)
+        research_run_id: Research-Run-Zuordnung
+        source_id: Quelle, von der das Audio stammt
+        segment_type: Art des Audio-Materials
+        transcript_text: Gesamtes Transkript als Text
+        audio_file_url: URL der Audio-Datei (optional)
+        duration_seconds: Gesamtdauer in Sekunden
+        language: Sprache des Audios
+        confidence: Confidence der STT-Erkennung
+        created_at / updated_at: Zeitstempel
+    """
+
+    __tablename__ = "audio_transcripts"
+
+    id = Column(String(36), primary_key=True, default=lambda: str(uuid4()))
+    research_run_id = Column(String(36), nullable=False)
+    source_id = Column(String(36), ForeignKey("sources.id"), nullable=True)
+
+    segment_type = Column(Enum(AudioSegmentType), nullable=True)
+
+    transcript_text = Column(Text, nullable=False, default="")
+    audio_file_url = Column(Text, nullable=True)
+    duration_seconds = Column(Float, nullable=False, default=0.0)
+    language = Column(String(16), nullable=False, default="unknown")
+    confidence = Column(Float, nullable=False, default=0.5)
+
+    created_at = Column(DateTime, nullable=False, default=datetime.utcnow)
+    updated_at = Column(DateTime, nullable=False, default=datetime.utcnow)
+
+    # Relationships
+    segments = relationship(
+        "AudioTranscriptSegmentModel",
+        back_populates="transcript",
+        cascade="all, delete-orphan",
+    )
+    claims = relationship(
+        "AudioClaimModel",
+        back_populates="transcript",
+        cascade="all, delete-orphan",
+    )
+
+    __table_args__ = (
+        Index("ix_audio_transcripts_research_run_id", "research_run_id"),
+        Index("ix_audio_transcripts_segment_type", "segment_type"),
+        Index("ix_audio_transcripts_source_id", "source_id"),
+    )
+
+
+class AudioTranscriptSegmentModel(Base):
+    """Segment des Audio-Transkripts (Stage 11).
+
+    Felder:
+        uuid: Primärschlüssel (UUID)
+        transcript_id: FK zum AudioTranscript
+        start_time: Start-Zeitstempel in Sekunden
+        end_time: Ende-Zeitstempel in Sekunden
+        text: Transkribierter Text
+        speaker_id: ID des Sprechers
+        speaker_type: Kategorie des Sprechers
+        confidence: Confidence der STT-Erkennung
+    """
+
+    __tablename__ = "audio_transcript_segments"
+
+    id = Column(String(36), primary_key=True, default=lambda: str(uuid4()))
+    transcript_id = Column(
+        String(36),
+        ForeignKey("audio_transcripts.id"),
+        nullable=False,
+    )
+
+    start_time = Column(Float, nullable=False, default=0.0)
+    end_time = Column(Float, nullable=False, default=0.0)
+    text = Column(Text, nullable=False, default="")
+    speaker_id = Column(String(64), nullable=False, default="")
+    speaker_type = Column(String(64), nullable=True)
+    confidence = Column(Float, nullable=False, default=0.5)
+
+    # Relationships
+    transcript = relationship("AudioTranscriptModel", back_populates="segments")
+
+    __table_args__ = (
+        Index("ix_audio_transcript_segments_transcript_id", "transcript_id"),
+        Index("ix_audio_transcript_segments_speaker_id", "speaker_id"),
+    )
+
+
+class AudioClaimModel(Base):
+    """Claim extrahiert aus Audio mit Zeitstempel (Stage 11).
+
+    Jeder Claim aus Audio hat einen Zeitstempel im Original-Audio
+    und muss quellenverknüpft sein (Provenance-Pflicht).
+
+    Felder:
+        uuid: Primärschlüssel (UUID)
+        transcript_id: FK zum AudioTranscript
+        claim_text: Die extrahierte Behauptung
+        timestamp_start: Start-Zeitstempel im Original-Audio
+        timestamp_end: Ende-Zeitstempel im Original-Audio
+        speaker_id: ID des Sprechers
+        source_url: URL der Quelle (Provenance)
+        evidence_span: Zitat oder Textpassage
+        claim_type: Art des Claims
+        confidence: Confidence der Claim-Extraktion
+    """
+
+    __tablename__ = "audio_claims"
+
+    id = Column(String(36), primary_key=True, default=lambda: str(uuid4()))
+    transcript_id = Column(
+        String(36),
+        ForeignKey("audio_transcripts.id"),
+        nullable=False,
+    )
+
+    claim_text = Column(Text, nullable=False)
+    timestamp_start = Column(Float, nullable=False, default=0.0)
+    timestamp_end = Column(Float, nullable=False, default=0.0)
+    speaker_id = Column(String(64), nullable=False, default="")
+    source_url = Column(Text, nullable=True)
+    evidence_span = Column(Text, nullable=True)
+    claim_type = Column(String(64), nullable=True)
+    confidence = Column(Float, nullable=False, default=0.5)
+
+    # Relationships
+    transcript = relationship("AudioTranscriptModel", back_populates="claims")
+
+    __table_args__ = (
+        Index("ix_audio_claims_transcript_id", "transcript_id"),
+        Index("ix_audio_claims_claim_type", "claim_type"),
+        Index("ix_audio_claims_speaker_id", "speaker_id"),
+    )