PDB AUDIO RESTORATION — AUDIO BOOKS & SPECIAL TAPES v2
Date: 2026-11-10

Source: Dr-Peter-Beter-HQ Audio books.zip
35 source MP3s, each decoded completely; 44.1 kHz mono.
Total preserved decoded duration: 10:04:38.

CLEANUP FILTER (FFmpeg 7.1.5):
highpass=f=60:p=2,afftdn=nr=4:nf=-50:tn=1:ad=0.85:gs=10

High-pass: 60 Hz, two poles.
Mild adaptive FFT noise reduction; maximum reduction parameter nr=4 dB.
This parameter does not imply a measured 4 dB improvement in every file.
No global hum notch, treble boost, low-pass, de-essing or gate.
No silence removal, time stretch, speech synthesis or content edits.

LOUDNESS:
Full cleaned audio measured with FFmpeg ebur128 (mono, not dual-mono).
One fixed gain per file = minimum of:
  -19.0 minus measured integrated LUFS;
  -2.5 minus measured true peak;
  +6.0 dB maximum positive gain.
The peak constraint takes priority over the loudness target.
No dynamic compression or limiting.

INTERMEDIATE:
32-bit floating-point WAV, 44.1 kHz mono. Processing peaks may exceed 0 dBFS
before gain; floating-point storage avoids clipping at this intermediate stage.

EXPORT:
libmp3lame, 128 kbit/s CBR, 44.1 kHz mono; Xing/LAME timing metadata.
Original artist/composer/album/date/track metadata and attached artwork retained.
Restored title and processing comment identify the derivative copy.
Exact decoded source/clean/output sample counts match for every file.
All final MP3s decoded and measured in full; output peaks are at most
-2.9 dBTP in this batch.

A/B PREVIEW:
Five 30-second pairs; original versus cleaned floating-point source.
The same per-file fixed gain is applied to BOTH versions, then both are encoded
once to 128 kbit/s mono MP3. The original excerpt is therefore level-adjusted;
the original archive source itself is unchanged. Output seeking follows input
decoding so inaccurate old MP3 headers cannot displace the compared excerpts.

LIMITATIONS:
Technical QC is not a manual listening review. Some hiss and hum remain.
No claim is made that missing audio detail was recovered. The untouched source
MP3s remain the archive reference. Per-file metrics and SHA-256 fingerprints are
in restoration-manifest.json.

REPRODUCTION:
Requires Python 3 and FFmpeg/FFprobe with libmp3lame, afftdn, ebur128 and astats.
Place original extracted MP3s in a source directory, then run:
python3 restore_pdb_audio.py --sources /path/to/Dr-Peter-Beter-HQ --manifest batch-manifest.json --output /path/to/restored --work /path/to/work --jobs 4 --stage all

Documentation:
https://ffmpeg.org/ffmpeg-filters.html#afftdn
https://ffmpeg.org/ffmpeg-filters.html#ebur128-1