Back to Tango Restoration

Osvaldo Fresedo — Firulete (1922)

One of the noisiest tracks in our collection — a 1922 acoustic-era recording with severe surface noise. Compare the original, the AI-restored version, and several postprocessing combinations to hear what each option adds.

All postprocessed versions start from the same AI-restored base. The AI noise removal is identical in every version — only the postprocessing differs. Listen and pick your favorite.
Source

Original Recording

1922
Unprocessed acoustic recording with heavy shellac surface noise, limited bandwidth, and no bass below ~200 Hz.
AI Restored

AI Restored (no postprocessing)

Base
ML noise removal (Mol2 + Demucs). Mono output, normalized to -16 LUFS. This is what the AI produces before any postprocessing.
Postprocessing Options

+ Stereo

Haas stereo widening (100%). Creates spatial depth from the mono source — instruments feel more spread out.

+ Stereo + Compression

Adds gentle 1.5:1 dynamic compression. Evens out volume differences — better for dancing in a milonga.

+ Stereo + Compression + Exciter

Adds a harmonic exciter (1.0). Generates subtle high-frequency overtones that bring sparkle and presence to the treble range.

+ Stereo + Harmonics

Pitch-tracked harmonic restoration (medium). Detects fundamental frequencies and synthesizes missing overtones for richer instrument timbre.

Full Treatment

Recommended
Stereo (120%) + compression (1.5:1) + exciter (1.0) + harmonics (medium). A balanced combination of all enhancements.

Dance Floor Ready

Wide stereo (150%) + firm compression (1.8:1) + subtle exciter (0.5) + louder (-14 LUFS). Tuned for PA systems at milongas.