radio: diarization pipeline fixes, benchmark setup, test episode set

- Fix voice_profiler threshold bug (HOST label overwrote Unknown unconditionally)
- Audio preload optimization: single ffmpeg per episode, 149.5x realtime on 5070 Ti
- WavLM threshold raised to 0.85 (Mike 0.90-0.99, callers 0.46-0.83)
- Promo/bumper filter: weighted signature scoring, 42->27 clean Q&A pairs
- Text-only Q&A fallback for episodes with no CALLER diarization labels
- TRANSFORMERS_OFFLINE=1 to skip HuggingFace freshness checks
- Add diarize_2018.py for targeted re-run + FTS5 rebuild
- Add benchmark.py + BENCH_SETUP.md for GURU-BEAST-ROG (RTX 4090) comparison
- Commit 9-episode training diarization.json outputs
- Session log: 2026-04-27-diarization-pipeline.md

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

projects/radio-show/audio-processor/src/clip_extractor.py

@@ -0,0 +1,105 @@
+"""
+Audio clip extraction using ffmpeg.
+Cuts clips from original broadcast MP3s for use in Audition/Audacity.
+"""
+
+import subprocess
+from pathlib import Path
+
+from rich.console import Console
+
+console = Console()
+
+
+def extract_clip(
+    source_path: Path,
+    start: float,
+    end: float,
+    output_path: Path,
+    padding: float = 1.5,
+    fade_ms: int = 200,
+) -> Path:
+    """
+    Extract a clip from source_path between start and end seconds.
+    Adds padding on both sides and applies fade in/out.
+    Returns the output path.
+    """
+    source_path = Path(source_path)
+    output_path = Path(output_path)
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+
+    clip_start = max(0.0, start - padding)
+    clip_end = end + padding
+    duration = clip_end - clip_start
+
+    fade_s = fade_ms / 1000.0
+
+    cmd = [
+        "ffmpeg", "-y",
+        "-ss", f"{clip_start:.3f}",
+        "-i", str(source_path),
+        "-t", f"{duration:.3f}",
+        "-af", f"afade=t=in:st=0:d={fade_s},afade=t=out:st={duration - fade_s:.3f}:d={fade_s}",
+        "-q:a", "2",
+        str(output_path),
+    ]
+
+    result = subprocess.run(cmd, capture_output=True, text=True)
+    if result.returncode != 0:
+        raise RuntimeError(f"ffmpeg failed: {result.stderr[-500:]}")
+
+    return output_path
+
+
+def extract_clips_for_results(results, output_dir: Path, padding: float = 1.5) -> dict[int, Path]:
+    """
+    Extract clips for a list of QAResult or SearchResult objects.
+    Returns {index: clip_path}.
+    """
+    output_dir = Path(output_dir)
+    clip_paths = {}
+
+    for i, result in enumerate(results):
+        episode = result.episode_id
+        audio_path = Path(result.audio_path)
+
+        if not audio_path.exists():
+            console.print(f"[yellow]Audio not found: {audio_path}[/yellow]")
+            continue
+
+        # Determine time range
+        if hasattr(result, "question_start"):
+            # QAResult
+            start = result.question_start
+            end = result.answer_end
+        else:
+            # SearchResult
+            start = result.start
+            end = result.end
+
+        def fmt(s):
+            m, sec = divmod(int(s), 60)
+            h, m = divmod(m, 60)
+            return f"{h}h{m:02d}m{sec:02d}s" if h else f"{m}m{sec:02d}s"
+
+        clip_name = f"{episode}_{fmt(start)}.mp3"
+        clip_path = output_dir / clip_name
+
+        try:
+            extract_clip(audio_path, start, end, clip_path, padding=padding)
+            clip_paths[i] = clip_path
+            console.print(f"[green]Clip {i+1}:[/green] {clip_name}")
+        except Exception as e:
+            console.print(f"[red]Clip {i+1} failed:[/red] {e}")
+
+    return clip_paths
+
+
+def format_timestamp(seconds: float) -> str:
+    """Format seconds as H:MM:SS or M:SS."""
+    h = int(seconds // 3600)
+    m = int((seconds % 3600) // 60)
+    s = int(seconds % 60)
+    if h:
+        return f"{h}:{m:02d}:{s:02d}"
+    return f"{m}:{s:02d}"