AES Journal Forum

Digital Audio Antiquing-Signal Processing Methods for Imitating the Sound Quality of Historical Recordings

(Subscribe to this discussion)

Document Thumbnail

Processing modern audio files to make them appear to have originated from historic technology, called antiquing, is the inverse of restoring old recordings. Simulating both global and local degradations of older technologies can be useful in such applications as testing restoration algorithms. A theoretical analysis of old recordings combined with numerous case studies results in a convincing imitation of historic phonograph and gramophone recordings. This work was motivated by a museum that wanted to show how audio technology has evolved over the last 150 years.

JAES Volume 56 Issue 3 pp. 115-139; March 2008
Publication Date:

Click to purchase paper as a non-member or you can login as an AES member to see more options.

(Comment on this paper)

Comments on this paper

Default Avatar
George Brock-Nannestad

Comment posted September 2, 2008 @ 17:40:23 UTC (Comment permalink)

Dear Editor,

We have thoroughly studied the paper published in the March 2008 issue of JAES by Välimäki et al. "Digital Audio Antiquing - Signal Processing Methods for Imitating the Sound Quality of Historical Recordings" [1]. This is a very interesting and well-written as well as well-documented paper, and the sound examples made available on the website are honest and without discussion the best version that can be made available: they are all wav. files. The paper also comes at an opportune moment - as a sort of memento of digital vs. analog in view of the 60 years anniversary of AES.

The need for digital signal processing methods in the museum exhibition field is very clear, although the authors do not really make this point. The traditional museal way of demonstrating the development of sound recording and reproduction has been by playing original recordings on original functioning replay equipment and making a studio recording of these and provide a switch in the exhibition area for use with headphones or loudspeakers. However, this restricts the examples to the selections made available for the historical replay equipment. By using digital signal processing, it is possible to take any modern recording and listen to it as it would hypothetically have sounded if recorded and reproduced on historical equipment. This is a very important approach, because it would permit an efficient manner of demonstrating both the historical development and the social differences between the types of reproduction equipment found in wealthy vs. less endowed homes, and completely without the wear on the original recordings that cheap constructions might entail.

However, this is where the approach as well as the results presented here break down in practice. The audible results have at most a superficial resemblance to historical distortions, and the well-documented paper makes us understand precisely why that is so. The authors have not analysed the historical evidence as regards the development of replay equipment and certainly not regarding the historical recording methods. We shall point to several sources of information that would have informed the project better and which would not have lead to misrepresentation of early audio recording and reproduction [2].

In essence, the approach by Välimäki at al. has been that of looking at the processes of sound restoration from historical recordings (to which one of the authors has already meritoriously contributed), at simulations of in particular guitar amplifier distortion, and at various digital sound effects schemes. However, one would have thought that it should have been the intention to model a recording-reproduction chain from various periods. And it is truly surprising that the authors did not avail themselves of the considerable know-how available in Finnish sound archive circles as to the sound of original equipment for comparison purposes.

The input to historical recording equipment were historical recordings, and the linear acoustical signal presented to the historical recording horns (i.e. pre-1925) or microphones had already been transformed by the recording process when it reached the grooves of a record. This has not been taken into account at all. Admittedly, sound restorers have rarely considered how the noises in early recordings were generated; nevertheless papers in AES publications exist that discuss the generating phenomena in detail. The narrower bandwidth of the original recording equipment also affected the inherent noise spectrum - this is not considered in the paper, and it contributes to making the sound examples unrealistic.

The only linear distortion Välimäki et al. consider for the acoustic phonograph and acoustic gramophone is that of the reproducing horn. But no account is taken of the Eigenresonances of the diaphragm and pressure-to-velocity transformation of the soundbox. The sound-box itself is driven by an elastic stylus in series with an elastic stylus bar. The neck of the horn never has a diameter of 6.3 cm, but rather in the order of 1 cm, because the hollow tonearm leading from the sound-box to the visible outer horn is also part of the horn. This setup was modelled by Maxfield and Harrison in 1925. Measurements on historical gramophones could have been performed by means of frequency test records surviving in the original shellac-based material to drive the sound-boxes and pickup of the time for present-day measurement in an anechoic chamber. The contemporary literature obviously already does provide some of this information. Finally, the acoustic reproduction in the 1930s via the high-end EMG handmade gramophone with a huge exponential horn was considered to be of a much higher standard than electrical gramophone-amplifier-loudspeaker consoles in the same price-range.

Each of the components of a recording-reproduction chain has an influence on the signal, but these are not discussed in relevant detail. For instance, the loudspeaker did not always use a dynamic motor, but earlier the better class used a so-called balanced-armature motor - these have very different characteristics both as to frequency range and distortion. The magnetic configuration and the corresponding distortions would also be found in cutter-heads (well into the 1950s) and pickups (in their heavy form until 1948). The launching of the Long Playing record in 1948 had several consequences, one of which was that there was no longer any coarse-groove research and development, and the published results from the 1950s and onwards concerning pickups are not directly transferable to earlier situations.

The rumble break-through in early cylinder phonographs and in gramophones had different sources. In cylinders it would frequently be caused by distortions of the surface, but it would not be audible in historical practice (although electrical reproduction for use with re-issues has to remove this disturbance). In gramophones it could typically be out of round of the so-called idler wheel between the motor shaft and the turntable, and this was cyclic and audible even in contemporary reproduction, but very distinct from the "thump".

Any eccentricity corresponding to a non-centered hole in a record would be discovered during manufacture (in large record companies 1 in 10 records were typically checked by playing them; small, local, manufacture may be different), and an example such as that provided by the authors would never have left a large factory. For this reason alone it is irrelevant to provide such a large distortion to the catalogue.

The major distortion not at all covered by the authors is the end-of-side distortion for a gramophone record. This is caused by the halving of the linear speed with respect to the outer radius when coming close to the label, and it is caused by the finite size of the needle now being in the same range as the radius of curvature of the waveform. There is in other words a gradual increase in the tracing distortion from beginning to end. And this becomes very audible.

The total distortion shown in Fig. 6b is typical, not of gramophone reproduction, but of cross-over distortion in a push-pull (class B) amplifier (which would normally apply at low levels!) simultaneously with saturation/almost clipping (which would normally apply at high levels!). From 1930-50 most amplifiers were single-ended, with a transformer, and would only display the saturation part. Push-pull (with vacuum tubes) was only used for more expensive amplifiers; with transistor amplifiers it became more common. The latter also suffered from Transient Inter-Modulation distortion, another characteristic distortion.

The click representation also merits some corrective comment. The characteristic sound in real life is caused by the fact that the pickup element was in many cases electro-dynamic, i.e. its response was the differential of the traced waveform. A particle would give rise to a peak of one polarity followed by a peak of the opposite polarity, with an amplitude that was inversely proportional to the length of the particle. This tended to overload the input stage in the phono preamplifier, which was usually only given a dynamic range corresponding to the music content, with some headroom. The sound heard on the examples is far from this reality.

The choice of a harpsichord as a reference signal is well-considered as regards providing a difficult but easily recognizable signal. But have any of the authors ever heard a harpsichord reproduced from a 1930s 78 rpm record, let alone from an acoustic recording? If recordings at the time had sounded as awful as the present examples, the industry would not have moved forward; the contemporary support that was expressed in record reviews in specialized magazines like the American Music Lover or the Gramophone (London) would never have materialized, but they would have concentrated on broadcast radio programmes instead.

All in all we feel sorry for the museums and their visitors that use this kind of aural demonstration, because it amounts to no more than free imagination, and it is a fake, like presenting a vase as an original, when in reality only 10% are original fragments, the rest being in-fill. It would be a pity if Prof. Perry R. Cook's term "antiquing" became derogatory.

May we suggest that space be made in the Journal for a historically informed paper about the phenomena that need to be simulated?

Peter G. Adamson, United Kingdom
George Brock-Nannestad, Denmark, Member
Dr. Peter Craven, United Kingdom, Member
Dr. Thomas Going, United Kingdom

P.S. It was not here the place to point out in detail how Välimäki et al.'s brief historical overview of the recording-reproduction chain suffers from the habitual errors and thereby perpetuates various myths.

[1] V. Välimäki, S. Gonzáles, O. Kimmelma, J. Parviainen, "Digital Audio Antiquing - Signal Processing Methods for Imitating the Sound Quality of Historical Recordings", JAES Vol. 56, No. 3, pp. 115-139 (March 2008)
[2] If the present comment is accepted as a Letter to the Editor, a proper list of references will be given.

Subscribe to this discussion

RSS Feed To be notified of new comments on this paper you can subscribe to this RSS feed. Forum users should login to see additional options.

Join this discussion!

If you would like to contribute to the discussion about this paper and are an AES member then you can login here:

If you are not yet an AES member and have something important to say about this paper then we urge you to join the AES today and make your voice heard. You can join online today by clicking here.

AES - Audio Engineering Society