Digitizing the Cylinders
In order to transfer the cylinders to digital format and subsequently the web, the Library's Archeophone, a specially made machine designed by Henri Chamoux, was used. The Archeophone is a universal cylinder player that uses electrical reproduction and modern styli (such as Stanton or Shure cartridges) to play back any of the varieties of cylinders made including 2 and 4 minute standard, intermediate and concert cylinders. The Archeophone allows for very fine control over the playback speed, correction of eccentricity, playback at half speed, and when used concurrently with the computer program Sound Forge (or another digital audio editing software program), transferring a cylinder to the digital .wav format is usually straightforward and quick. Correcting eccentricity is the most time consuming portion of the process. All cylinders were played back on the Archeophone, using either a two minute stylus or a four minute stylus as appropriate. Cylinders were played at standard Edison speeds, typically 160 for black wax and Blue Amberol cylinders. Brown wax cylinders were pitched by ear and usually ended up being recorded at 140rpm. Actual cylinder recording speeds may vary and thus our playback speeds may not be entirely accurate. However, attempting to correctly pitch each individual cylinder would not only be incredibly time consuming, but would probably not result in a greater percentage of accurately pitched cylinders. Each cylinder was captured as a .wav file on the library's Digital Audio Workstation. Some transfers sound better than others, but after consulting with Chamoux we feel that a wider variety of styluses to match with individual cylinders would be helpful.
Compressing the Audio for Web Delivery
After recording the cylinders as a .wav file, we chose a set of standard file formats that would satisfy the need for durability and posterity, ubiquity and public accessibility in making the files available over the web. While we could deliver uncompressed audio over the web, with modern compression technology there is really no need. Selecting a compression codec (a codec, or COmpression DECompression algorithm reduces the size of files) was initially somewhat daunting, due to the great number of audio file formats and compression codecs in existence and the potential of the technologies becoming obsolete in a few years. After careful consideration, we chose two file formats for web delivery of audio--mp3 and Quicktime. We ruled out Realmedia and Windows Media for two reasons: both require streaming servers to deliver streaming content on the web and the files can't be converted back to editable formats. Streaming servers are necessary for projects with many simultaneous users such as streaming audio course reserves or webcasting of radio which may have dozens or hundreds of simultaneous users. For delivery of digitized audio in a library context with a few users accessing many different files at different times, we believe a streaming server would be an added expense and maintenance issue and it would be unlikely that hundreds of people would be listening to these files simultaneously. Windows Media and Realmedia also are not fully compatible across platforms like Quicktime and the files can't be "decoded" back into files like .wav or other editable files. While Microsoft and others do this to protect copyright holders, we felt it was unacceptable for a project using public domain material where people have a right to do whatever they want with the audio. Quicktime also offers the use of a streaming server, but also allows progressive download, where the file begins to play as soon as a portion of the file has begun to download. We chose the progressive download format and in most cases it will give users the appearance of streaming, though there is no streaming server to manage the flow in the case of network problems which Realserver and Quicktime server both do. With progressive download Quicktime files, the only server needed is a http server (a regular web server) which the library already maintains.
We wanted a downloadable file that people could load onto their computers, and mp3 was chosen for its near ubiquity for distribution of non-streamed audio. For streamed audio we tested 7 different compression codecs. As it turned out, the mp3 compression codec turned out to be by far the best compromise between file size and quality for streaming audio as well. As a result, we also chose mp3 compression for the Quicktime files. In the process of testing we made about 250 test files at different bit rates, sampling rates, and with different codecs. Some of the more revealing examples from our tests concerning the various compression codecs are offered below. We were surprised that some of the highly lauded codecs such as the QDesign codec, did not handle the noise inherent in the cylinder recordings and produced unacceptable artifacts in the compressed audio.
In compressing the files from .wav format we have used Cleaner 5 software to compress the files. For mp3 file compression we have chosen the Fraunhoefer mp3 Codec at a sampling rate of 22.5 kHz with a data rate of 56 kbits/sec. Through extensive comparison, we found that there was nothing to be gained by using higher sampling rate (cylinders just don't contain any musical information over 11Khz) and the 56 kbits/sec data rate produced a file of acceptable quality. Higher bit rates such as 96, 128 or even 256 which are clearly beneficial when encoding modern stereo recordings proved to be of no value in encoding the cylinder recordings.
For the streaming audio format, two Quicktime file format options were made for those who want instantaneous playback: a larger, higher-quality broadband version in Quicktime format using an mp3 codec at a sampling rate of 22.5 kHz with a data rate of 56 kbit/sec (the same as the mp3 format file), and a smaller, lower-quality 56k modem dial-up connection version in Quicktime format using an mp3 codec at a sampling rate of 16 kHz at a data rate of 20 kbits/sec. We found this file would stream over a 56K modem connection, with fairly good quality.
Additionally, we set up Cleaner 5 to normalize the audio at 96%, and encode them on the slowest setting even though we could detect no difference between the slower and faster encoding of the files. The files were encoded as batches overnight so we didn't care if the computer took longer to compress the files! Maybe somebody else could tell the difference even if we couldn't. However, there was a noticeable decrease in quality using the fastest compression speed.
The listener may find the sound quality on some of the cylinders to be quite poor at times, especially with the brown wax-cylinders. Although we have tried to select both interesting and high-quality cylinders, it is important to keep in mind that in comparison to modern mediums like LPs or CDs, or even electrically recorded 78s, the cylinder format had its limitations and furthermore, time has not always been kind to many cylinders. Certainly, there are a variety of noise reduction methods, both hardware and software, but we found that when applying them to our cylinder collection, it not only reduced the background noise and hiss, but took out some of the music as well and in some cases left artifacts. Therefore, the cylinders presented are essentially unaltered from their original sound. If we had access to Cedar noise reduction equipment or another system of sufficiently high quality, we would have made at least one cleaned up and de-noised file for web delivery.
Without cataloging the collection, there is no point in digitizing it. People need to have an access point to the bibliographic information on the cylinders. Item level cataloging is an important component of UCSB's program for managing our collections of historic audio recordings. Syracuse University's Belfer Audio Archive has cataloged many of their cylinders and while there is overlap between the collections, a large percentage of our collection is unique. We used and expanded on their source copy whenever possible. We upgrade cataloging to the standards set for cataloging of historical sounds recordings at UCSB, including complete tracings for composers and performers, uniform titles when necessary and subject headings. All cylinders digitized for this project are cataloged in Pegasus, the UCSB Libraries' online catalog. In the advanced search mode, the patron can limit their search to just cylinder recordings.
Additionally, links are provided in the catalog record which take the library patron directly to the audio file. One of the largest deterrents to using a collection of historical sound recordings such as this is that because of the fragility of the media, patrons do not handle the recordings and library staff have to make a copy of each requested item. For all but the most dedicated or desperate researchers, this is usually too much work. Library patrons are increasingly becoming used to online journals, e-books and electronic indexes. To have to use a physical object, let alone one that may require a wait of several days to be copied and access, is becoming too much for the patron who simply wants to hear a particular style of music, or an old pop song, or an example of early 20th century performance practice.
Selected samples of audio files compressed using Cleaner 5
Test of mp3 at various sampling and bit rates
- 44Khz mono, 96 kbit/sec (2731K) CUSB-CYL0016a-MT1.mp3
- 44Khz mono, 96 kbit/sec 11 Khz low pass (2731K) CUSB-CYL0016a-MT1c.mp3
- 44Khz mono, 96 kbit/sec 15 Khz low pass (2731K) CUSB-CYL0016a-MT1d.mp3
- 44Khz mono, 128 kbit/sec (3638K) CUSB-CYL0016a-MT2.mp3
- 44Khz mono, 128 kbit/sec 11 Khz low pass (3638K) CUSB-CYL0016a-MT2c.mp3
- 44Khz mono, 128 kbit/sec 15 Khz low pass (3638K) CUSB-CYL0016a-MT2d.mp3
- 44Khz mono, 192 kbit/sec (5461K) CUSB-CYL0016a-MT3.mp3
- 44Khz mono, 192 kbit/sec 11 Khz low pass (5461K) CUSB-CYL0016a-MT3c.mp3
- 44Khz mono, 192 kbit/sec 15 Khz low pass (5461K) CUSB-CYL0016a-MT3d.mp3
- 16 Khz mono, 24 kbit/sec (700K) CUSB-CYL0013a-MT16.mp3
- 22.5 Khz, mono, 48 kbit/sec (1393K) CUSB-CYL0013a-MT22.mp3
- 24 Khz mono, 56 kbit/sec (1632K) CUSB-CYL0013a-MT24.mp3
- 32 Khz mono, 96 kbit/sec (2798K) CUSB-CYL0013a-MT32.mp3
- 24 Khz mono, 48 kbit/sec (1425K) CUSB-CYL0008-24-48.mp3
- 22 Khz mono, 56 kbit/sec (1654K) CUSB-CYL0008-22-56.mp3
- 22 Khz mono, 48 kbit/sec (1418K) CUSB-CYL0008-22-48.mp3
- 24 Khz mono, 56 kbit/sec (1662K) CUSB-CYL0008-24-56.mp3
Quicktime file with Qdesign codec
- 11Khz mono, 12 kbit/sec CUSB-CYL0025-QD56.mov