Draft:INA-Saphir

INA-Saphir is a process that optically retrieves sound from a disc phonograph record. It was initiated by INA in 2002.

Introduction
Disc phonograph record, invented before 1889 by Emile_Berliner, were the main recording and playback medium from the extinction of Phonograph_cylinder (1929) to the generalization of Tape_recorder (early 1950s). The easiest way of recovering the sound contents of a disc record is usually playing it on a turntable using a physical stylus, but a number of conditions may render the record unplayable physically. The most difficult records are often unique lacquer recordings, broken, or with cracked or delaminating lacquer. Optical playback is often the only solution for recovering the sound of those damaged records.

History
The INA-Saphir process was imagined in 2002 by Louis Laborelli in INA, patented in 2004, and described in articles in 2007.

The project was developed as a collaboration between INA and Indeep between 2007 and 2011. Since then, the project was developed principally in INA under the Saphir or INA-Saphir name.

The first recoveries of otherwise unplayable records were made in 2016.

The first external uses of the system were in March 2019 where a number of records were scanned in Montreuil for Gecko, October-November 2019 in Hilversum where 36 sides were scanned  for the participants to the 50th IASA conference and JTS 2019 hosted by  Beeld en Geluid, and in 2021 in  Montreuil again, where 51 disc sides were scanned and decoded by Gecko for  VRT.

Scanning process
As the ELP_Japan Laser_turntable system, INA-Saphir uses the reflection of light to measure the slope of the groove with respect to the tangential direction. This allows the highest frequencies to be recovered.

A specifically-designed condenser casts rays with varying colours onto a 2.5x2mm region of the record, at 45 degrees, i.e. perpendicularly to the groove walls. Rays reflected are collected through a doublet lens pair and focused on a board 2D camera sensor.

The full active surface of the record is scanned as 40 to 70 rings of 1250 overlapping colour pictures, stored as raw files.

Image Processing and Sound Extraction
The Saphir Play software tool extracts from each picture the successive track segments, and converts the colour to a cutter velocity measurement i.e. an audio signal value.

The resulting segments are stored on a map of the disc side, and overlapping segments are fused into fragments.

The resulting fragments are connected in the correct order for generating the sequence of audio samples. The audio samples are stored in a wav audio file.

Broken and Cracked Records
When the scanned record is broken or cracked, or when pieces are missing, the correct order for playing the audio fragments has to be decided. This is done using a shortest path solver, penalising long jumps, but with reward (negative costs) for using as much as possible of the available fragments.

Operator involvement is usually necessary at this step, to add constraints that will help :
 * guiding the preferred connections
 * cutting fragments at cracks crossings
 * solving the problem faster

System quality
The audio quality obtained from the INA-Saphir system is quite dependent on the condition of the record. In the best conditions, on a pristine vinyl 78rpm record, THD+N was measured at 3.1% (-30dB) in the 20Hz–20kHz band.

The quality is usually worse than obtained from a proper mechanical stylus playback. Therefore the main usage of the INA-Saphir system is considered to be the preservation of records that cannot be played mechanically.

Advantages and disadvantages
The approach used by INA-Saphir provides a set of advantages over conventional or optical disc records playback:


 * Can scan and recover records in very poor condition
 * Scans are usually made under a 3mm glass pane, that protects the record and keeps it flat
 * Up to the highest useful frequencies can be recovered (20kHz at 78rpm)
 * Compact: scanner size is smaller than a conventional disc turntable
 * Can skip over any crack/break/hole/corruption
 * Large range of disc records playability, including lateral and vertical cut records

Disadvantages of the system:
 * Slow : on average 30min per scan, 1 hour for decoding, up to several hours of effort for reconstruction of very damaged records
 * Intermittent signal acquisition can generate in artefacts that have to be filtered out
 * Signal quality worse than conventional playback
 * Only applicable to disc recordings

Recovered sample files
First recording by fado singer Maria_Teresa_de_Noronha: Fado de Mouraria (1939). Aluminum-based lacquer disc, numerous fine cracks. 

From INA collections, Set W1425, "Voici ma carte Monsieur : L'existentialisme Pictural !", 10 minutes live recording by Pierre_Sabbagh of the first French post-war (spade) duel, 1946. Zinc-base lacquer disc, severely degraded, very brittle lacquer, peeling off. 

From CNRS Centre for Research in Ethnomusicology collections, ethno-musicological recordings by  Maurice Leenhardt in New Caledonia (1939). Zinc-based lacquer discs, 25cm, severely damaged (multiple delaminated lacquer flakes, exudates). 

From VRT collections, first 65 seconds of "Comité Tegen Het Alcoholisme Antwerpen" in, and "Peter Benoit - Reprise Van Scherzo Uit Strijkkwartet Op.10" , earliest known  VRT recording of string quartet Op. 10 by composer Peter_Benoit.