ffmpeg-skill 2.19.0 → 2.20.0

package/docs/contract.md

@@ -21,7 +21,7 @@ The contract is derived from the code that runs, not maintained beside it:
 | Field | Meaning | Changes when |
 |---|---|---|
 | `contract_version` | shape of this document (`1.0`) | a key is renamed, removed or changes meaning |
-| `skill.version` | the npm / package.json version (`2.19.0`) | any release |
+| `skill.version` | the npm / package.json version (`2.20.0`) | any release |
 
 A release that adds a tool or a flag keeps `contract_version`; a breaking change to the
 ToolSpec shape bumps it. Consumers pin on `contract_version` and read `skill.version`
@@ -131,7 +131,7 @@ JSON would have been rewritten for that. The flat keys, typed per tool in `outpu
   "deprecated": [{"what": "running on Python 3.9 (the requirements.python floor)", "since": "2.10.9", "replacement": "Python 3.10 or later; ...", "removed_in": "3.0.0", "where": "behaviour"},
                  {"what": "running on FFmpeg 5.x (the requirements.ffmpeg floor)", "since": "2.14.0", "replacement": "FFmpeg 6.1 or later; ...", "removed_in": "3.0.0", "where": "behaviour"}],
   "removed": [{"what": "...", "since": "1.10.0", "replacement": "...", "removed_in": "2.0.0", "where": "cli | json | mcp | behaviour"}],
-  "skill": {"id": "ffmpeg-skill", "version": "2.19.0", "execution_mode": "local", "kind": "execution",
+  "skill": {"id": "ffmpeg-skill", "version": "2.20.0", "execution_mode": "local", "kind": "execution",
             "entrypoints": {"cli": "...", "mcp": "...", "contract": "...", "doctor": "..."},
             "not_provided": ["AI reasoning", "decisions", "production plans", "project IR", "approvals", "network access", "transcription engine"]},
   "requirements": {"python": ">=3.9 (standard library only; 3.9 deprecated since 2.10.9, removed in 3.0.0; 3.10+ recommended)", "ffmpeg": ">=5.0 (5.x deprecated since 2.14.0, removed in 3.0.0; 6.1+ recommended)", "ffprobe": ">=5.0"},
@@ -699,6 +699,16 @@ from where the expected value came, and what was not measured.
 | `audio_substituted` | `join.py` (video join) | the clips that had no audio stream, whose span of the join is generated silence; `notes` says the same |
 | loudness `tolerance.tp` | `loudness.py`, `export.py`, and `check.py`'s true-peak row | the one dBTP margin every tool applies (`_platforms.TP_TOLERANCE_DB`, 0.1 dB); before 2.18 it was 0.1, 0.05 and 0.0 in the three tools |
 
+Keys added in 2.20, all additive, from the same paper's baseline experiment (each was a length or a
+silence a tool reported as fine while another measurement disagreed):
+
+| key | tool | what it holds |
+|---|---|---|
+| `verification[]` step `duration` | `export.py` (a preset with a platform maximum, without `--allow-long`) | the written file's length against the maximum, measured as `check.py` measures it (the container's duration, `tolerance` 0). A length that cannot be read fails the step with `reason: not_measured` |
+| `verification[]` step `duration` (`property: audio_duration`) | `loudness.py` | the output's sound against the input's, to one AAC frame (at least 0.03 s): a levels pass changes no timing. A length that cannot be read fails the step with `reason: not_measured` |
+| `inputs_duration`, `transition_shortening`; step `duration`'s `inputs_total` | `join.py` | the clips' own total and how much the transitions took out of it. `expected_duration` is derived from the transition the join applied, so it cannot show that on its own |
+| `silence_kinds`, `silence_breakdown` | `silence.py` | per detected silence, `digital` (at or below -90 dBFS) or `below_threshold` (quiet sound under `--threshold`), and the count and seconds of each with the level that defines it |
+
 Per-tool keys added after 2.10.11 ("nothing happens silently"), all additive:
 
 | key | tool | what it holds |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ffmpeg-skill",
-  "version": "2.19.0",
+  "version": "2.20.0",
   "description": "Agent Skill that gives coding agents (Claude Code, Cursor, Codex) a local video editor: 42 FFmpeg tools with a machine-readable contract, contract-derived MCP server, FFmpeg capability detection, probe-first / verify-last workflow. Cut, join, silence removal, fit, captions and karaoke, overlays, motion graphics, HDR to SDR, LUTs, audio clean-up and typed dynamics, sync with drift correction, multicam, loudness, delivery checks, project rendering, batch. No API keys, no cloud, no dependencies.",
   "keywords": [
     "ffmpeg",

package/scripts/_common/probe.py

@@ -300,6 +300,9 @@ def probe(path: str, role: str = "input") -> Dict[str, Any]:
             "bitrate": _to_int(audio.get("bit_rate")),
             "duration": _to_float(audio.get("duration")),
             "start_time": _to_float(audio.get("start_time")),
+            # Matroska/WebM give no stream duration, only a DURATION tag; the container's own length
+            # is the longest stream's, which is not the sound's when the picture runs longer (2.20)
+            "tagged_duration": _tag_seconds((audio.get("tags") or {}).get("DURATION")),
         }
         # every audio stream in file order: index n here is `-map 0:a:n` (audio.py --audio-stream n)
         out["audio_streams"] = [{
@@ -497,6 +500,19 @@ def analyze_levels(path: str, seconds: float = 20.0) -> Dict[str, Any]:
     }
 
 
+def _tag_seconds(value: Any) -> "Optional[float]":
+    """A Matroska DURATION tag ("00:00:04.008000000") in seconds, or None."""
+    try:
+        if "-" in str(value):  # "-00:00:01": int("-00") is 0, so the sign would be lost
+            return None
+        h, m, sec = str(value).split(":")
+        seconds = int(h) * 3600 + int(m) * 60 + float(sec)
+    except (TypeError, ValueError):
+        return None
+    # float() reads "nan" and "inf": a length that is not a finite, non-negative number is none
+    return seconds if math.isfinite(seconds) and seconds >= 0 else None
+
+
 def _to_float(v: Any) -> Optional[float]:
     try:
         return float(v) if v is not None else None

package/scripts/export.py

@@ -333,6 +333,20 @@ def main() -> int:
     elif platform and has_audio and args.normalize:
         spec = PLATFORMS[platform]
         notes.append(f"[dry-run] --normalize: loudness.py -I {spec['lufs']:g} --tp {spec['tp']:g} would run on the written file if it misses {platform}'s spec")
+    if p["max"] and not args.allow_long and not STATE.dry_run:
+        # the preset's length promise, measured the way check.py measures it (the container's
+        # duration, no tolerance): before 2.20 nothing re-read the length, and a --normalize pass
+        # that left the sound 0.1 s long shipped a 90.10 s Reels file as verified
+        written = probe(output, role="output").get("duration")
+        if written is not None:
+            step = {"step": "duration", "property": "duration", "ok": written <= p["max"],
+                    "expected": {"max": p["max"]}, "measured": round(written, 3),
+                    "delta": round(written - p["max"], 3), "tolerance": 0.0, "provenance": "specification"}
+        else:
+            # the preset promised a length; one that could not be read is not one that held
+            step = {"step": "duration", "property": "duration", "ok": False, "expected": {"max": p["max"]},
+                    "reason": "not_measured: the written file reports no duration"}
+        extra.setdefault("verification", []).append(step)
     if notes:
         extra["notes"] = notes
     # the delivery keeps the source's picture and, except as a GIF, its sound

package/scripts/join.py

@@ -81,6 +81,20 @@ def unpending_length(durs: List[float], d: float) -> Optional[float]:
     return round(sum(durs) - d * (len(durs) - 1), 3)
 
 
+def overlap_report(durs: List[float], expected: Optional[float]) -> Dict[str, Any]:
+    """The clips' own total next to the joined length (2.20). The joined length is derived from
+    the transition the join itself applied, so a check against it alone cannot show that a fade
+    took time out of the material; the sum of the inputs is the yardstick the user holds."""
+    if expected is None or STATE_PENDING:
+        return {}
+    total = round(sum(durs), 3)
+    out: Dict[str, Any] = {"inputs_duration": total, "transition_shortening": round(total - expected, 3)}
+    if out["transition_shortening"] > 0.0005:
+        STATE_NOTES.append(f"the transitions overlap the clips: {fmt_secs(total)} of clips play as {fmt_secs(expected)} "
+                           f"({out['transition_shortening']:.3f} s shorter)")
+    return out
+
+
 def clip_length(meta: Dict[str, Any], fps: float) -> float:
     """The one length a clip holds in a join, for its picture and its sound alike: the
     video stream's, or the audio stream's when the sound runs more than a frame past the picture
@@ -175,7 +189,8 @@ def join_audio(args: argparse.Namespace, metas: List[dict]) -> int:
             die(f"{output} is {a.get('sample_rate')} Hz {a.get('channels')} ch, expected {rate} Hz {channels} ch")
     info(f"wrote {output} ({fmt_secs(r['duration'])}, {expected_text(expected)}, audio {a.get('codec')} {channels}ch {rate}Hz, {n} clips, "
          + ("crossfade" if d else "butt join") + ")")
-    emit(output, mode="audio", **reported(), clips=n, transition=args.transition if d else "none", expected_duration=expected,
+    overlap = overlap_report(durs, expected)
+    emit(output, mode="audio", **overlap, **reported(), clips=n, transition=args.transition if d else "none", expected_duration=expected,
          sample_rate=rate, channels=channels, video=False)
     return 0
 
@@ -568,7 +583,7 @@ def main() -> int:
         verification.append({"step": "duration", "property": "video_duration", "ok": abs(measured - expected) <= tol,
                              "measured": round(measured, 3), "expected": expected,
                              "delta": round(measured - expected, 3), "tolerance": round(tol, 3),
-                             "provenance": "input_premeasurement"})
+                             "provenance": "input_premeasurement", "inputs_total": round(sum(lens), 3)})
     # a clip with no audio stream is joined over generated silence: the join is right to do that, but
     # the report must say which clips' sound is not theirs (the paper's join finding: absence was
     # filled and nothing in the result showed it)
@@ -577,7 +592,8 @@ def main() -> int:
         STATE_NOTES.append(f"{len(substituted)} clip(s) had no audio stream; their span of the join is generated "
                            "silence: " + ", ".join(os.path.basename(p) for p in substituted))
     info(f"wrote {output} ({fmt_secs(measured)}, {expected_text(expected)}, {w}x{h} @ {fps:g}fps, {n} clips, {args.transition})")
-    emit(output, mode="video", **reported(), clips=n, transition=args.transition, expected_duration=expected,
+    overlap = overlap_report(lens, expected)
+    emit(output, mode="video", **overlap, **reported(), clips=n, transition=args.transition, expected_duration=expected,
          verification=verification, expect_streams={"video": True, "audio": True},
          audio_substituted=substituted,
          dropped_non_av_streams=any(m.get("subtitle_streams") or m.get("data_streams") for m in metas))

package/scripts/loudness.py

@@ -20,6 +20,8 @@ import re
 import sys
 
 from _platforms import TP_TOLERANCE_DB
+
+AUDIO_LENGTH_TOLERANCE = 0.03  # seconds: one AAC frame at 48 kHz is 0.021 s
 from _common import STATE, add_common, apply_common, emit, AUDIO_CODECS, audio_codec_for, default_output, die, ffmpeg_base, info, probe, require_tool, run, run_analysis, run_keeping_subtitles, dry_run_input_pending
 
 
@@ -93,6 +95,12 @@ def main() -> int:
     bitrate = args.audio_bitrate or "192k"
 
     dropped_streams = False
+    # loudnorm emits its last block whole, and through the AAC encoder that came out as 0.1 s more
+    # sound than the input had (a 90.000 s Reels export left at 90.100 s, which check.py fails):
+    # the sound is trimmed back to the input's own length. `duration=` counts from the first sample,
+    # so sound that starts late keeps its end (`end=` is a timestamp, and cut it). atrim follows
+    # loudnorm directly: an aresample between them fails channel-layout negotiation on FFmpeg 5.1
+    in_audio = audio_length(meta)
 
     def encode(tp: float, bitrate: str) -> None:
         nonlocal dropped_streams
@@ -101,6 +109,8 @@ def main() -> int:
             f":measured_I={stats['input_i']}:measured_TP={stats['input_tp']}:measured_LRA={stats['input_lra']}"
             f":measured_thresh={stats['input_thresh']}:offset={stats['target_offset']}:linear=true:print_format=summary"
         )
+        if in_audio:
+            af += f",atrim=duration={float(in_audio):.6f}"
         cmd = ffmpeg_base() + ["-i", args.input, "-af", af, "-ar", str(sr)]
         if ext in AUDIO_CODECS or not meta.get("video"):
             cmd += ["-vn"] + audio_codec_for(output, bitrate)
@@ -158,19 +168,40 @@ def main() -> int:
             f"the encoder overshoots more than the loudnorm ceiling can absorb at this bitrate",
             kind="verification", output=output, result=result,
             hint="raise --audio-bitrate (e.g. 256k) or deliver a lossless format (wav/flac) and let the platform encode")
+    verification = [{"step": "loudness",
+                     "ok": bool(after.get("silent")) or (abs(float(after["input_i"]) - args.lufs) <= 1.0 and float(after["input_tp"]) <= args.tp + TP_TOLERANCE_DB),
+                     "lufs": float(after["input_i"]), "tp": float(after["input_tp"]), "target_lufs": args.lufs, "target_tp": args.tp,
+                     "property": "loudness", "expected": {"lufs": args.lufs, "tp_max": args.tp},
+                     "measured": {"lufs": float(after["input_i"]), "tp": float(after["input_tp"])},
+                     "delta": {"lufs": round(float(after["input_i"]) - args.lufs, 2), "tp": round(float(after["input_tp"]) - args.tp, 2)},
+                     "tolerance": {"lufs": 1.0, "tp": TP_TOLERANCE_DB}, "provenance": "user_specified"}]
+    written = probe(output, role="output")
+    out_audio = audio_length(written)
+    if in_audio and out_audio:
+        # a levels pass changes no timing: the sound is as long as the input's, to one AAC frame
+        tol = max(1024.0 / float(sr), AUDIO_LENGTH_TOLERANCE)
+        verification.append({"step": "duration", "property": "audio_duration", "ok": abs(out_audio - in_audio) <= tol,
+                             "expected": round(float(in_audio), 3), "measured": round(float(out_audio), 3),
+                             "delta": round(float(out_audio) - float(in_audio), 3), "tolerance": round(tol, 3),
+                             "provenance": "input_premeasurement"})
+    else:
+        # a length that could not be read is not a length that matched
+        verification.append({"step": "duration", "property": "audio_duration", "ok": False,
+                             "reason": "not_measured: the " + ("input" if not in_audio else "output") + " reports no duration"})
     emit(output, result=result, dropped_non_av_streams=dropped_streams,
          measured={k: stats[k] for k in ("input_i", "input_tp", "input_lra", "input_thresh", "target_offset", "silent")},
-         targets={"lufs": args.lufs, "tp": args.tp, "lra": args.lra},
-         verification=[{"step": "loudness",
-                        "ok": bool(after.get("silent")) or (abs(float(after["input_i"]) - args.lufs) <= 1.0 and float(after["input_tp"]) <= args.tp + TP_TOLERANCE_DB),
-                        "lufs": float(after["input_i"]), "tp": float(after["input_tp"]), "target_lufs": args.lufs, "target_tp": args.tp,
-                        "property": "loudness", "expected": {"lufs": args.lufs, "tp_max": args.tp},
-                        "measured": {"lufs": float(after["input_i"]), "tp": float(after["input_tp"])},
-                        "delta": {"lufs": round(float(after["input_i"]) - args.lufs, 2), "tp": round(float(after["input_tp"]) - args.tp, 2)},
-                        "tolerance": {"lufs": 1.0, "tp": TP_TOLERANCE_DB}, "provenance": "user_specified"}])
+         targets={"lufs": args.lufs, "tp": args.tp, "lra": args.lra}, verification=verification)
     return 0
 
 
+def audio_length(meta: dict) -> "Optional[float]":
+    """The sound's own length: the stream's duration, else Matroska's DURATION tag. Never the
+    container's, which is the picture's when the picture runs longer (and would fail a correct
+    output against an expected length the sound never had)."""
+    audio = meta.get("audio") or {}
+    return audio.get("duration") or audio.get("tagged_duration")
+
+
 def _kbps(value: str) -> int:
     v = value.lower().rstrip("k")
     try:

package/scripts/silence.py

@@ -15,7 +15,7 @@ import argparse
 import json
 import os
 import sys
-from typing import List, Tuple
+from typing import Any, Dict, List, Tuple
 
 # `detect` moved into _common/probe.py in 1.16.0 so metadata.py --auto-chapters can measure the
 # same silences without importing this tool; the body is unchanged and the name still lives here.
@@ -44,6 +44,7 @@ def merge_spans(spans: List[Tuple[float, float]]) -> List[Tuple[float, float]]:
     return out
 
 
+DIGITAL_SILENCE_DB = -90.0  # 2.20: at or below this a span is digital silence (16-bit dither sits near -96)
 BREATH_FLOOR = 0.12  # shortest gap silencedetect is asked for under --speech-aware
 
 
@@ -255,6 +256,33 @@ def _words_from_transcript(data):
     return out, had_segments, untimed
 
 
+def silence_kinds(path: str, silences: "List[Tuple[float, float]]", duration: float,
+                  threshold: float) -> "Dict[str, Any]":
+    """Which detected silences are digital silence and which are quiet sound under the threshold
+    (2.20). Both are removed alike, but they are not the same finding: a room tone, a breath or a
+    whisper under --threshold is sound, and a report that called all of it "silence" hid that the
+    cut took audible material. One more silencedetect pass at DIGITAL_SILENCE_DB finds the digital
+    spans; a detected silence that is at least 99% digital counts as digital."""
+    if not silences or STATE.dry_run:
+        return {}
+    digital = [(s, duration if e == float("inf") else e) for s, e in detect(path, DIGITAL_SILENCE_DB, 0.02)]
+    labels, tally = [], {"digital": [0, 0.0], "below_threshold": [0, 0.0]}
+    for s, e in silences:
+        e = duration if e == float("inf") else e
+        span = max(e - s, 1e-9)
+        covered = sum(max(0.0, min(e, de) - max(s, ds)) for ds, de in digital)
+        kind = "digital" if covered / span >= 0.99 else "below_threshold"
+        labels.append(kind)
+        tally[kind][0] += 1
+        tally[kind][1] += e - s
+    return {"silence_kinds": labels,
+            "silence_breakdown": {"digital": {"count": tally["digital"][0], "seconds": round(tally["digital"][1], 3),
+                                              "level_db": DIGITAL_SILENCE_DB},
+                                  "below_threshold": {"count": tally["below_threshold"][0],
+                                                      "seconds": round(tally["below_threshold"][1], 3),
+                                                      "level_db": threshold}}}
+
+
 def main() -> int:
     ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
     ap.add_argument("input")
@@ -344,6 +372,7 @@ def main() -> int:
     if args.filler:
         silence_keeps = keep_ranges(merge_spans(list(silences)), duration, args.margin, args.min_keep)
         silence_only = max(0.0, duration - sum(e - s for s, e in silence_keeps))
+    kinds = silence_kinds(args.input, silences, duration, args.threshold)
     summary = {
         "silences": [[round(s, 3), None if e == float("inf") else round(e, 3)] for s, e in silences],
         "keep": [[round(s, 3), round(e, 3)] for s, e in keeps],
@@ -351,6 +380,8 @@ def main() -> int:
         "kept_duration": round(kept, 3),
         "removed_seconds": round(silence_only, 3),
     }
+    if kinds:
+        summary.update(kinds)
     if args.speech_aware:
         summary["speech_aware"] = {
             "min_silence": args.min_silence, "floor": min(BREATH_FLOOR, args.min_silence),