[Table of Contents]


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ARSCLIST] IRENE article



I must wholeheartedly agree with Marcos. In many ways the IRENE version sounds inferior, the definition of her voice and the overall intelligibility of the vocal seems considerably worse and the piano sounds like there is a pillow over the instrument. However, the fact that the record was not played and further degraded by the process is obviously a breakthrough from a preservation perspective. I am sure that by tweaking the algorithm further that the clearly superior qualities of the stylus version will be incorporated as the technology matures. What I am most impressed with was the surprisingly excellent resolution of the original
acoustical recording made in 1923...subtract the noise artifacts and it sounds a lot like an MP3 from iTunes carved in 2007.


Aaron


Marcos Sueiro wrote:
Unfortunately, the sound samples in the article are not doing IRENE a great favour: Sure, there is much less crackle, but the level of the IRENE sample is significantly lower (rendering the S/N ratio probably about the same), and it has much higher wow.

Neat idea, nevertheless, and one with great potential.

Marcos

Farris Wahbeh wrote:
Here's a nice article on IRENE published in UC Berkeley's "Science Matters"
journal.
The link to the article has neat sound scan pictures and links to sound
samples: http://sciencematters.berkeley.edu/archives/volume4/issue30/story1.php


More info on IRENE is here: http://irene.lbl.gov.

Rescuing Recorded Sound from Silence

by Kathleen M. Wong

Researchers Carl Haber and Vitaliy Fedayev of Lawrence Berkeley National
Laboratories working on IRENE. credit: LBNL

While listening to National Public Radio in 2000, Carl Haber learned that
the Library of Congress had a big problem. The Library's audio collection,
which spans the 130-year history of recorded sound, includes the soaring
tenor of Enrico Caruso, the speeches of Teddy Roosevelt, and the voices of
Native Americans from now-vanished tribes. These echoes of a bygone era were
recorded on media such as wax cylinders and shellac and lacquer discs. But
many are now too fragile to play in their original format; the pressure of a
stylus or phonograph needle could cause irreversible damage. Others are too
broken, worn or scratched to yield high-quality sound. The archivists needed
a means to preserve the recordings without injuring them further.


A physicist with Lawrence Berkeley National Laboratory (LBNL), Haber was
developing subatomic particle detectors to be used at CERN in Geneva,
Switzerland. This involved using digital cameras and robots to place each
delicate detector in precisely the right place. In a flash of insight, Haber
realized that an optical scanning system could solve the Library's quandary.


Millions of historical sound recordings such as this wax cylinder are in
need of preservation in the United States alone. credit: courtesy Carl Haber


"I had phonograph records as a kid, so I knew sound was stored in a
mechanical profile. I realized that we could use images to figure out in
detail what the groove actually looked like, and use a computer to calculate
the sound. I thought that might be a way to get around the problem of things
being delicate and damaged; you wouldn't have to touch them," Haber says.


Haber already had access to a machine that could make high-resolution
digital scans. Postdoctoral fellow Vitaliy Fadeyev wrote a computer program
to control the turntable and translate the images into sound.


Haber used a narrow beam of light to illuminate the record's surface. The
flat bottoms of the grooves and the spaces between tracks appeared white;
the sloped sides of the grooves, scratches and dirt looked black. The image
was then analyzed by computer. The program found the edges of each groove by
focusing on areas of high contrast. It could correct areas where scratches,
breaks or wear made the groove wider or narrower than normal.


A digital scan of phonograph grooves taken by IRENE. The side-to-side
wiggles of the groove contain the audio information. credit: Carl Haber

That first test was agonizingly slow. Forty minutes of scanning was required
to obtain just one second of audio. But it provided what the scientists
needed-proof of principle. And the scan played far more cleanly and clearly
than the worn original disc.


Haber and Fadeyev wrote a paper describing the device and sent it,
unsolicited, to the Library of Congress. The next thing Haber knew, he had
an invitation to visit the Library to talk about the technique. By 2004,
Haber and Fadeyev were developing ways to scan discs and cylinders more
efficiently.


The two types of media presented very different problems. On antique
monaural discs, sound is recorded in horizontal wiggles of the record
groove. On cylinders, sound is recorded in the vertical plane-the depth of
the groove.


Millions of historical sound recordings such as this wax cylinder are in
need of preservation in the United States alone. credit: courtesy Carl Haber


"With discs, we used a camera to image them at high resolution in two
dimensions. Once we understood how cylinders were recorded, we realized we
had to measure the third dimension (3D) as well," Haber says.


In 2005, LBNL engineers Earl Cornell and Robert Nordmeyer joined the
project. With the Library's urging, the team concentrated on producing a
dedicated disc scanner. Dubbed IRENE (after the Weavers' "Good Night,
Irene," the first disc the team scanned), the device was installed at the
Library last summer for evaluation and needs just four seconds to scan one
second of audio.


The group is now refining a device that scans in 3D. The device is based
upon a type of confocal microscope. White light directed at the surface of a
cylinder or disc passes through a lens. But the lens is imperfect by design;
though it splits the light into its component colors, each color comes into
focus at a different depth. The color of the reflected light reveals the
height of the scanned point. The computer assembles these points into
profiles for each groove and translates the data into sound.


A digital scan of phonograph grooves taken by IRENE. The side-to-side
wiggles of the groove contain the audio information. credit: Carl Haber

The current 3D scanning process takes 20 hours to record one minute of
sound. But a new version of the confocal scanner, developed for the dental
industry, should reduce that to about 10 minutes.


A half-dozen physics and engineering undergraduates from UC Berkeley have
been instrumental in speeding the project along. "Students can apply the
kinds of techniques they learn in classes about statistics, mathematical
analysis and signal processing to a project they can really get their arms
around," Haber says. A Berkeley graduate student in linguistics is poised to
join the project later this summer.


UC Berkeley's Phoebe Hearst Museum and Native Americans are among those who
could benefit the most from IRENE and its sister 3D scanner. In the early
1900s, UC Berkeley anthropologist Alfred Kroeber and colleagues recorded the
legends, songs, customs and voices of dozens of California Indians on some
3,000 one-of-a-kind wax cylinders. Many of these tribes and languages have
since died out or are on the verge of extinction. The LBNL group is now
collaborating with linguist Andrew Garrett and Victoria Bradshaw of the
museum to digitize the Kroeber recordings. Remastering these cylinders could
help new generations of native peoples study their ancestral customs and
tongues—and help carry the sounds of the past into the future.
Sound Samples



[Subject index] [Index for current month] [Table of Contents]