Slashdot Mirror
From the Higgs Boson Particle to Leadbelly
Roland Piquepaille writes "Physicists from the Lawrence Berkeley National Laboratory are using the same methods to search for the elusive Higgs Boson particle and to digitally restore audio recordings from the past. Berkeley Lab signed an agreement with the Library of Congress to digitize the many thousands of early blues or jazz recordings it has in its archives. And the results are spectacular. Compare for example, these two versions of "Good Bye Irene", before and after being optically reconstructed (WAV format, 18 and 19 seconds). This news release describes the method used by the physicists. This overview contains other details and extra references about this project." We also covered finding Higgs Boson recently as well.
1). It's "Ledbelly"
2). It's "Good Night Irene"
This reminds me of this project (which has been Slashdotted before) which can be done with a home scanner. But this new Berkeley method is obviously much more advanced.
Back in the good old days of vinyl records, RIAA Equalization was/is an industry standard for how music that is recorded on vinyl records is played back. The idea is to compensate for the fact that vinyl does not have a flat audio frequncy response.
The link above explains it much better (and in more detail) that I can.
\/Don
The LoC collection of American folk music was certainly one of the strangest ventures ever carried out by the US government. In a way, it paralleled the ancient Chinese venture that resulted in the Shih Ching (Book Of Odes). Both govs sent recording agents into the country with the directive to collect the songs of the people. The Chinese had only ink & paper (or whatever they used for paper circa 800 BC), while their US counterparts (beginning in the 1920s, I think) utilized their day's equivalent of direct-to-disk recording, i.e., big in-field acetate disc cutters with acoustic recording gear. For the most part these intrepid researchers are unknown, but they collected an incredible mass of disparate music. Black & white music from the deep South and the Appalachians, cowboy music from the plains states, music from native American tribes... The impression I have is that they were told something like "Go ye forth, collect their songs so we may know the mind of the the people". Well, that's what the Chinese collectors did anyway...
There are some well-known LoC recordings that have gained some fame, including a series of recordings by Leadbelly and an awesome set of music and reminiscences by Jelly Roll Morton. However, both those sets were recorded "in studio" and are not field recordings. They are magnificent though.
Btw, I should make special mention of the Lomax family. Father John and son Alan were responsible for some remarkable recordings, including the work by Leadbelly and Jelly Roll. Alan also made the earliest recordings of Muddy Waters and some excellent recordings of Son House while working for the LoC. John was something of a Texas cracker (check out his dialog with Willie McTell on the LoC recordings), but he was a brave man going into some of the places he visited. He also wrote a very weird account of his acquaintance with Leadbelly in a book he wrote about the great self-proclaimed King of the 12-string Guitar..
Some of the catalog has been available to the public for quite a while, but I doubt that catalog has listed anything close to the amount of material the LoC must have in their vaults. Those acetate masters won't last forever, and I'm glad to learn that an attempt will be made to save those recordings.
Btw, I doubt copyright is an issue with this material. Unless I'm mistaken I believe all of it is in the public domain now. Perhaps someone else can clarify ?
No recent US administration would dream of doing such a project now. They definitely would *not* want to do it to know the collective mind of the people...
Result sound viewed with -42 threshold
Result sound viewed with -60 threshold
Result sound viewed with -42 threshold
And the following image is a spectrogram of the original "Goodbye Irene" file:
Original sound viewed with -42 threshold
Each of these spectrograms was computed using 1024 point Discrete Fourier Transforms with a factor of 8 overlap. The dimensions of the images are unlabeled, but provide a frequency range of 0Hz - 22050Hz along the vertical axis, and approximately 344 horizontal pixels represent one second of time. Darkness represents the magnitude of the signal at a particular measured frequency.
With significant interest, I can perhaps label these axes for easier reading. Simply keep in mind that the top of the vertical dimension represents 22050 Hz.
Given the sound quality of the result sound provided, utilizing 16-bit quantization with a sampling rate of 44.1Khz is more than adequate. But while the result is promising, it is hardly archival quality in my opinion, due to the obvious digital artifacts.
The dynamic range of this particular music is confined by musical convention and the microphone technology available for the recording. The theoretical 96dB of dynamic range availed by 16-bit quantization is more than sufficient to represent the dynamic range of this particular music (and many others) recorded with similarly early microphony and disc-cutting technology.
The frequency range of the music does not appear (in this result mind you) to have significant musical information above an approximate (but conservative) 11000 Hz. The frequency range availed by a sampling frequency of 44.1 KHz is more than adequate to quite faithfully represent this music. To significantly reduce the broadband pops and crackles in the recording, high frequency information is lost. Further, the recording technology available at the time probably could not accurately transduce such frequencies from the original performance either.
The spectrogram reveals that the undulating noise in the result sound occurs at a nearly precise 5Hz. It also reveals that this "noise" is obviously an artifact of the restoration process; it really isn't noise, but the result of a time-varying filter which cuts gaussian lobes into the spectrum of the music from approximately 4000Hz to 9200Hz in a manner somewhat a kin to a wah wah pedal. The lobes can be seen clearly in all of the spectrograms I provided, but they appear more stark as the peak threshold of the spectral plots decrease. Their duration is quite close to .05 seconds.
In my opinion, archives should preserve physical recording media as long as possible to allow transduction techniques to mature. I find the 5Hz filtering artifact present in this result to make the current state of this particular optical transduction process unacceptable for archiving. It would be a shame to replace physical media with music colored with such avoidable artifacts. I am sure that such artifacts can be alleviated and that optical scanning of phonograph records (discs and cylinders) has great promise as a transduction technique.