Sound Reproduction R & D Home Page


Using methods derived from our work on instrumentation for particle physics we have investigated the problem of audio reconstruction from mechanical recordings.  The idea was to acquire digital maps of the surface of the media, without contact, and then apply image analysis methods to recover the audio data and reduce noise.  This work has been described in two papers which have been published in the Journal of the Audio Engineering Society (J. Audio Eng. Soc., vol. 51, no. 12, pp.1172-1185 (2003 Dec.) (link), and J. Audio Eng. Soc., vol. 53, no.6, pp.485-508 (2005 June)) (link). The most recent status reports on this research were presented at various meetings in Fall/Spring 2006-2007 and can be downloaded here (contains many sound clips).

NEW!! June 2007 “Topics in Preservation Science Lecture” given at the Library of Congress, June 18, 2007, PDF version of talk (40 MB)

April 2007 Presentation at Stanford University Center for Computer Research in Music and Acoustics PDF version of talk (80 MB !!)

February 2007 Most Recent Project Status  PDF version of Project Status Presentation February 8, 2007 (12 MB)

Jan 2007 Library of Congress Meeting PDF version of talk presented at the Library of Congress, January 12, 2007 (7 MB)

Dec 2006  PDF version of talk presented at “New Journeys in Collaboration” the California Tribal Museum Partnership Summit, December 13, 2006 (25 MB)

Oct 2006  PDF version of talk presented to the October 2006 Meeting of the Audio Engineering Society San Francisco, talk contains many new results (66 MB)

Oct 2006 PDF version of talk presented to the UC Berkeley CITRIS Seminar, talk contains many new results (46 MB)


Below are presented various sound clips from early studies (2002-2004) of these methods as well as earlier presentations. In 2005 the National Endowement for the Humanities funded the development of a prototype scanning machine based upon this research. This machine, called IRENE, is currently under evaluation at the Library of Congress in Washington D.C. and is described in the aforementioned presentations above (see in particular the June 2007 file).


The work described here represents experiments and studies designed to explore the potential of certain optical methods to reconstruct and preserve mechanical recordings and aid in the process of mass digitization. The sound clips presented should not be regarded as a definitive final result, only as a proof-of-principle demonstrations or test of these optical methods. No attempt was made to add any digital post processing, such as additional noise reduction, unless explicitly stated here.

 

Two related approaches have been explored to date. In the first, described in detail in the first  JAES paper (pdf file), methods of two dimensional (2D) imaging were used to map a portion of the surface of 78 r.p.m. disc records. The 2D imaging focused on the groove bottom and sound was reconstructed from a measurement of its trajectory. A 2D approach is possible for pre-stereophonic discs because the audio is stored in lateral or side-to-side modulations of the groove path. The results of this study can be heard in the first and second sound clips below. In the second approach, described in the second paper (pdf file), a 3D surface measuring technique called confocal scanning microscopy, was used to map a portion of the surface an Edison cylinder. A 3D approach is required for cylinders because the audio is stored in the vertical or hill-and-dale modulations of the surface. The results of this study can be heard in the third sound clip below. The 3D methods are more powerful than the 2D methods because they allow the entire surface to be analyzed rather than a plane projection or slice. However the scan times for 3D imaging are, in general, considerably slower.


Additional information can also be found in the presentations.

 

First clip (2D method): "Goodnight Irene", written by H.Ledbetter and J.Lomax, performed by the Weavers (1950),78r.p.m. shellac disc

Sound clip reconstructed by optical method 1.7 Mbytes, 19.1 seconds

Same sound clip played from record with stylus and turntable

Same sound clip from recent CD version, remastered from original magnetic tapes

Optical version after (commercial) digital noise reduction (here additional digital noise reduction has been applied to the audio waveform)

Sound of optically read groove amplitudes before differentiation into velocites (this refers to a technical issued discussed in the JAES paper)

Second Clip (2D method): "Nobody Knows the Trouble I've Seen", traditional, performed by Marion Anderson (1947),78r.p.m. shellac disc

Sound clip reconstructed by optical method 1.1 Mbytes, 12 seconds

Same sound clip played from record with stylus and turntable

Same sound clip played from recent CD version, remastered from original studio source material

Optical version after (commercial) digital noise reduction (here additional digital noise reduction has been applied to the audio waveform)

Sound of optically read groove amplitudes before differentiation into velocities (this refers to a technical issue discussed in the JAES paper)

Third Clip (3D method): "Just Before the Battle, Mother", written by George F. Root, performed by Will Oakland and Chorus (1912), issued on an Edison Blue Amberol 4 minute celluloid cylinder. More information about this recording can be found here .

Sound clip reconstructed by optical method 29 seconds

Sound clip played from cylinder with modern electrical stylus 29 seconds

Optical version after (commercial) digital noise reduction (here additional digital noise reduction has been applied to the audio waveform)


23-March-2005: Fourth Clip (2D method with dedicated hardware): "The Star Spangled Banner", performed by Kate Smith, issued on a shellac 78 rpm disc Victor 26198-B. This clip is presented also to illustrate the absence of modulation in the background noise as was heard on the earlier "Goodnight Irene" and "Nobody Knows the Trouble I've Seen" clips, identified as a resolution effect. .

Sound clip reconstructed by optical method 40 seconds

Sound clip played from shellac disc with modern electrical stylus 40 seconds

Optical version after (commercial) digital noise reduction (here additional digital noise reduction has been applied to the audio waveform)


Additional details can be found in the JAES papers and in presentations given at the Library of Congress and other venues.The documents from 2003 pre-date the 3D work of the third sound clip but do discuss the general 3D imaging issues. The documents from 2004 discusses the third sound clip. The documents from 2005 discuss new results on high speed 2D scanning (the IRENE project) and new 3D results including early work on damaged cylinders.

PDF version of original paper describing this work, submitted to JAES and published December 2003

PDF version of second paper describing reconstruction of an Edison cylinder  submitted to JAES, and published June 2005

PDF version of talk presented in the Lawrence Berkeley National Lab Instrumentation seminar series 2006, (15 MB)

 

PDF version of talk presented at the University of Toronto and University of Rochester, physics colloquia 21-22-Sept-2005, (33 MB)

PDF version of talk presented at the Library of Congress 16-May-2005, (33 MB)

PDF version of talk presented at the Lawrence Berkeley National Lab Summer Lecture Series 29-May-2005, (43 MB)

Powerpoint (ppt) version of Library of Congress presentation 28-May-2004

PDF version  of presentation at LBNL Physics Division Research Progress Meeting September 2003

Powerpoint (ppt) version of Library of Congress presentation 18-June-2003

Quicktime movie (11 MB) showing scanning system at Berkeley.  First shot shows confocal probe scanning along the axis of a Blue Amberol cylinder.  Second shot is at a lower angle with light spot (barely) visable.  Third shot shows the PC screen with a portion of the measured groove profile.

Movie

Still photographs of the scanning system

 View 1
 View 2
 

Movie explaining sound reconstruction process produced by Berkeley Lab Technology Transfer Division for R&D100 (approx. 10 minutes long)

Movie
 


Sponsors of this research are the Library of Congress, The Mellon Foundation, The National Endowement for the Humanities, The National Archives and Records Administration, The John Simon Guggenheim Memorial Foundation, The University of California at Berkeley, and the US Department of Energy.


E-mail me at chhaber@lbl.gov