Putting the Byte on Noise--NoNoise from Sonic Solution (Mar. 1989)

Home | Audio Magazine | Stereo Review magazine | Good Sound | Troubleshooting


Departments | Features | ADs | Equipment | Music/Recordings | History





by Michael Wright

Noise. Nasty interference that might he music to the ears of a John Cage but is tune like Public Enemy Number One to audiophiles. For years, audio engineers have worked assiduously to keep this sonic miscreant on the run, improving hardware and wielding noise-reducing weapons with names like Dolby and dbx.

Enter the digital revolution, with its expanded frequency response and, more important, dramatically lowered noise floor. Record companies, scrambling to rush out catalogs in the new CD format, came face lo face with the old nemesis attain. Tape hiss, clicks, and pops stood ready to blare out of the background silence. How to fight back?

There's a promising new weapon in the arsenal of the Sound Police that uses computer treatment to de-noise the legacy of primitive recording techniques and deteriorated masters which hampers our enjoyment of a rich musical heritage. Called NoNoise by its inventors at San Francisco's Sonic Solutions, it's a method that's gaining rapid acceptance among producers.

You might say this new technology is another offspring of Star Wars (the movie, not the defense initiative. Sonic Solutions was founded in 1986 by Robert Doris, President; Mary Sauer, Vice President of Marketing and Operations; and James A. Moorer, Vice President of Audio Development, who were all involved with digital systems for film and television post-production at the Lucasfilm subsidiary, The Droid Works.

What is NoNoise? In a nutshell, it uses very sophisticated digital signal-processing software-combined with human interaction-to de-noise digitized audio data (usually) derived from analog sources.

Prior to receiving NoNoise treatment, the best analog source material available--anything from old 78-rpm records to shellac, metal, glass, acetate, multi-track tape, or even optical film soundtrack masters-is transferred by the record company to digital audio tape, usually the Sony PCM-1630 format, the current standard for CD master tapes. This digital audio data is then loaded via a special interface from the Sony PCM-1630 processor onto two very large 700-megabyte computer disk drives. Because digital audio is so data intensive, the storage requirements are enormous. In the Sony 1630 format, for example, there are 44,100 16-bit samples per second, requiring almost 11 megabytes of storage for each minute of recorded material. The NoNoise disk drives can hold roughly one hour of stereo music. If this amount of data were in the form of a written text, it would be 500,000 pages long! Essentially, NoNoise is a two-phase process that combats two categories of noise: Transient or impulsive clicks and pops, and recurrent or ongoing hiss, surface noise, rumble, buzz, and hum. To achieve the best results, Sonic Solutions generally prefers to work with a flat transfer from the original recording. Equalization applied before de-noising usually brings up background noise, so it is typically performed after removal of tape hiss or background noise in order to give the producer more flexibility in applying EQ. The NoNoise process begins with a visual analysis or diagnosis, which involves browsing through graphic waveform displays of the music presented in both time and frequency domains. This allows the operator to evaluate the extent of noise pollution and to identify areas which might present special problems. This review is repeated after treatment.

De-noising then begins with work on transient noise. Until NoNoise, the primary technique for eliminating clicks and pops involved editing them out of the program, leaving a minuscule gap-anywhere from 5 to 10 mS -- in the music signal. Usually, this is closed up, slightly altering the piece's timing.

NoNoise transient processing involves not only removing the noise, but actually re-synthesizing the waveform as well. Clicks and pops appear as spikes or plateaus on the waveform graphs (Fig. 1). Parameters are selected, and the NoNoise software automatically seeks out the aberrations and performs microsurgery. Once the offending noise is excised, the software employs artificial-intelligence techniques to sample the waveform areas surrounding the operative site. It then digitally reconstructs the sound beneath the click, using estimated projections based on the sampling of the surrounding waveforms (Fig. 2). More difficult than excising transient noises, however, is the problem of lowering the noise floor-reducing recurrent hiss and background or surface noise. This is made even more complicated by the fact that noise frequencies usually coexist with program signal. It is at this point that the gray areas of aesthetics, subjective producer judgment and experience, and record company objectives also enter into the equation.

Without NoNoise, the primary techniques for handling continuous noise included re-equalization (especially rolling off high-end frequencies and boosting others), notch filtering, and broadband analog noise gating. Depending on how heavily they're used, these methods almost always result in a significant alteration of both the harmonic content and original intent of the music.

NoNoise de-hissing begins by seeking out as pure a sample of noise as possible. Ideally, this sample would be taken from a place without any program signal, such as at the head before the music begins, at the tail after a fade, or in a caesura between phrases.

If no pure sample is available, the best noise sample possible is selected and its frequency spectrum edited by an operator to eliminate music or program content. This noise sample then becomes a noise "fingerprint" which is unique to each recording and is the critical element in the de-noising program.

At this point, the producer must decide how heavily the de-noising is to be applied. "Treatment can be light, medium, or heavy, depending on the noise level of the source and what the producer wants," explains Mary Sauer.

"The heavier the treatment, the more the potential impact on the music." Tests are run to see how various parameters affect the music. Since these tests are conducted on perfect digital "copies" residing on the computer drives, variations may be spun out until the desired result is achieved.

In essence, what follows is a very sophisticated form of digital filtering directed by artificial intelligence. The audio data is divided into more than 2,000 very precise frequency bins.

Each of these is analyzed, in turn, by the computer, using an elaborate algorithm which compares the sonic energy level of the program signal in each bin to the noise fingerprint.


above: James A. Moorer, Sonic Solutions' Vice President of Audio Development, at the keyboard of the company's Macintosh-based work station.

If the signal energy level in a given frequency bin is greater than the fingerprint's parametric threshold, the signal is left untouched. If signal energy level is less than the noise fingerprint, an attenuation factor is applied to reduce the noise. Figure 3 shows a waveform display of an untreated piece of music and a close-up view of the energy spectrum between 11 and 12 kHz, where very little program signal, but much noise, is present. Figure 4 shows the same waveform and energy view after treatment. The music signal is virtually identical and intact (except for.-the removal of noise spikes). The energy spectrum/from 11 to 12 kHz, however, shows that the average energy of the peaks has shifted down about 13 dB, from -92 to -105 dB. This analysis and filtering is done continuously throughout the piece of music; during de-hissing, more than 53 million separate calculations will be performed on each second of sound! Complete final processing of the master is accomplished automatically, usually overnight. De-noising one hour of music takes roughly 8 to 10 hours. A de-noised digital master tape is made, and the record company can then begin its final prep work (editing, mixing, equalization, etc.).

That's how NoNoise processing works. How well it works is another question. As with situational ethics, the answer is, "It depends."

--------------


Fig. 1-Music waveform with two clicks and a dropout, from film soundtrack of Jim Morrison and The Doors, Live At The Hollywood Bowl.

Fig. 2-Same waveform as Fig. 1 after de-clicking and reconstruction using I the NoNoise process. The black bar shows where the signal has been I reconstructed.

Fig. 3-Waveform of noisy signal (A) and close-up view of its energy spectrum from 11 to 12 kHz (B). For this signal, the energy in that spectral region is almost all noise.

Fig. 4-Effects of NoNoise treatment on the signal of Fig. 3. Note that removal of the noise spikes from the waveform (A) has also reduced its amplitude, as shown by the scale at left, but has left its overall shape unchanged. The amplitude loss is even more marked (again, note the left-hand scale) in the noise-ridden spectral region from 11 to 12 kHz (B).

---------------

First of all, in terms of the elimination of transient noises and signal reconstruction, NoNoise is a stunning success. The most dramatic example of this kind of treatment can be heard in the reclamation of the audio portion of The Doors' 1968 concert film Live At The Hollywood Bowl, a CD version of which has been released by Elektra (60741-2). At the beginning of the performance, Jim Morrison knocked loose the mike lead running to the mobile recording unit. The PA system was unaffected, so the concert proceeded; it took engineers almost 15 minutes to isolate and correct the problem. The resulting noise rendered the footage useless until NoNoise was brought to bear, removing the noise caused by the loose connection and restoring Morrison's vocals. This saved an otherwise irretrievable event from oblivion. On the CD, you cannot tell that any of this happened.

Perhaps not as spectacular, but no less dramatic, have been the restorations performed on the RCA Bluebird CD releases of quintessential jazz from Duke Ellington, Bix Beiderbecke, and many others whose original masters have experienced substantial deterioration over the last 40 or 50 years.

"Until we discovered NoNoise, we could not realistically produce CDs of performances from the pre-tape era, a cornerstone of the RCA jazz catalog," says executive producer Steve Backer. "Our goal was to remove the noise without distorting the music."

"For the typical song in the RCA project, we removed around 800 clicks," adds Sonic Solutions production manager John Polito, "and in some tracks, as many as 3,500." A good example can be heard on Jelly Roll Morton's The Pearls (Bluebird 6588-2-RB), which still exhibits clicks and pops on some tunes but is vastly superior to the original.

The results of NoNoise treatment were presented in A/B comparisons at a recent press audition, and the contrast was truly impressive.

You can perhaps simulate the effect if you have access to any of the original 78-rpm records and can play them side by side with the CD.

Another example worth noting is from Philips' Legendary Classics series of recordings, many of which rely heavily on NoNoise processing (as discussed by Bert Whyte in "Behind the Scenes," August 1988). Because the masters were just too noisy, Philips had held back on CD releases of many of these historically significant recordings-dating back as far as 1928-by Casals, Monteux, Richter, and others.

Specifically, you can hear the results on the Dvorák Concerto in B Minor for Cello and Orchestra, Opus 104, with Emanuel Feuermann and the National Orchestral Association under Leon Barzin, recorded in New York in 1940 (Philips CD 420 776-2). While again you'll have to imagine the difference--unless you have an original-the master for the CD has been remarkably restored. You will still hear a great deal of surface noise, but the improvements are amazing. Fortunately, Philips produced a press-only CD which included NB snippets from the series, so I have been able to compare the results.

Even though the treatment has been expensive-as much as $6,000 for one CD-Philips is pleased with the results.

"We strove to keep the signal as close as possible to the original music, affecting only the noise itself," explains Taadatsu Atarashi, Vice President of Marketing of Philips Classics. Indeed, Philips is so pleased, it has licensed the use of NoNoise in Europe and Japan, hoping to establish a universal standard.

However, when evaluating the de-hissing process--removing recurrent noise--certain limitations of NoNoise treatment become apparent. To be fair, this is a very complex situation because these limitations are partly inherent, partly technological, and partly due to subjective factors which include the values and experience of the producer.

Issues such as type of music, instrumentation, and condition of the original aside, the primary inherent problem is simply that noise and music coexist within the same frequency ranges. And as long as the music is to be preserved, this noise cannot be entirely removed. Even with vastly greater computer processing power than is currently available, performing much finer sonic microsurgery (a costly proposition at this point in time) and eliminating noise through subtraction can never be absolute. In technological terms, this problem may never be overcome unless we're willing to allow digital reconstruction techniques far more complex than already discussed.

Certainly, in frequency ranges where no program signal is present, NoNoise is extremely effective. This can be easily observed in treated recordings. Listen, for example, to the 1962 recording of Stravinsky's L'Histoire du Soldat (Philips CD 420 773-2), featuring Jean Cocteau and Peter Ustinov under Igor Markevitch's direction. In the extended, unaccompanied dramatic passages, the noise reduction is particularly effective.

Also, in music where the program signal is maintained at a fairly constant level, either because of denser orchestration (e.g., symphonic music) or style (e.g., rock or be-bop), NoNoise performs splendidly.

Check out the Warner Bros. reissue of The Grateful Dead's Europe '72 (2668-2) for a good example of material that was recorded under less than ideal conditions but still generally sounds excellent after NoNoise treatment, with just a smidgen of hiss audible.

The real problem arises with music which is sparsely orchestrated (e.g., solo instruments) or has a very wide dynamic range. As long as the signal overpowers the noise, your ear is likely to ignore distractions such as hiss. But in quiet passages or as notes decay, the presence of hiss can actually seem exaggerated. This results in a disturbing "noise pumping" effect which can be quite disconcerting.

===========================

De-Noised Discs

There is a partial list of titles processed by Sonic Solutions and NoNoise.

In most cases, the entire master was processed; however, on some discs (notably The Beatles'), not all songs were de-noised. No track-by-track information is available from record companies. Although some labels, such as Philips and RCA, publicize use of NoNoise and feature the logo on treated recordings, others are very shy about admitting use of any treatment. M. W.

Jazz & Blues Recordings Louis Armstrong, Pops: The 1940s Small-Band Sides (Bluebird 6378-2-RB). Classic combos; clean and consistent throughout.

Sidney Bechet, The Legendary Sidney Bechet (Bluebird 6590-2-RB). Recordings from 1932 to 1941 with various groups; amazingly clean and consistent; remarkable job.

Duke Ellington, The Blanton-Webster Band (Bluebird 5659-2-RB, three CDs). Super-smooth Duke from 1939 to 1942; clean and clear all the way through, with only a hint of noise at higher volumes.

Duke Ellington, Black, Brown & Beige (Bluebird 6641-2-RB, three CDs). A tad more hiss and surface noise than on The Blanton-Webster Band and from more diverse sources; still excellent, quintessential Duke from 1944 to 1946.

Duke Ellington, The Great Ellington Units (Bluebird 6751-2-RB). Extremely low hiss; mixed surface noise depending on the source; excellent 1940 to 1941 anthology.

Paul Horn, Inside (Rykodisc RCD-10040). Classic jazz event primitively recorded inside the Taj Mahal in 1969; greatly reduced tape hiss, although still audible at louder volumes.

Jelly Roll Morton, The Pearls (Bluebird 6588-2-RB). Hot classics given new life; noise floor dramatically reduced but occasionally inevitable surface noise due to 78-rpm sources.

Various, At The Jazz Band Ball: Chicago/ New York Dixieland (Bluebird 6752-2RB). Northern versions of 1929 to 1939 Southern jazz; remarkably clean and noise free.

Various, Classic Jazz Piano-1927-1957 (Bluebird 6754-2-RB). Very little hiss, occasional surface noise; otherwise, a superb, consistent-sounding collection.

Various, Great Trumpets: From Jazz to Swing (Bluebird 6753-2-RB). Great blowing from 1927 to 1946; remarkably clear and consistent, with a minimum of hiss at very high volumes.

Various, The Metronome All-Star Bands (Bluebird 7636-2-RB). The best jazz artists from 1937 to 1949; generally clean, with low hiss and irregular low surface noise depending on source.

Classical Recordings Pablo Casals, Beethoven: Piano Trios No. 7 ("Archduke") and No. 5 ("Ghost") (Philips CD 420 855-2). Cello maestro in live recordings; No. 7 (from 1958) has slight hiss, and No. 5 (from 1961) is very clean.

Jean Cocteau and Peter Ustinov (actors) and Igor Markevitch (conductor), Igor Stravinsky: L'Histoire du Soldat (Philips CD 420 773-2). A 1962 recording with excellent presence; minimal noise floor remains after consistent de-hissing.

Pierre Monteux and Concertgebouw Orchestra of Amsterdam, Beethoven: Symphony No. 3 ("Eroica") (Philips CD 420 853-2). Great performance, great sound from 1962.

Maurice Ravel and Serge Prokofiev, Ravel Conducts Ravel/Prokofiev Conducts Prokofiev (Philips CD 420 778-2). Definitive pieces of mixed quality depending on source; "Bolero" hissy; "Chansons Madécasses" clear but surface noise; "Romeo and Juliet, Suite No. 2" moderately noisy.

Sviatoslav Richter, Mussorgsky: Pictures at an Exhibition; Schubert; Chopin; Liszt (Philips CD 420 774-2). Remarkably clean 1959 live concert with noise floor audible as volume increases.

Andrés Segovia, The Segovia Collection, Vols. 2 & 3 (MCA Classics MCAD42067 and MCAD-42069). Legendary performances marred by prominent hiss and noise-pumping effects.

Gérard Souzay, Schubert: Die Schbne Mullerin (Philips CD 420 850-2). Extremely low noise floor for sparkling 1964 performance by legendary baritone with piano accompaniment.

Rock/Pop Recordings The Andrews Sisters, 50th Anniversary Collection, Vol. 1 (MCA MCAD-42044). Walk down Memory Lane with hits from 1937 to 1950; varying levels of surface noise still present.

The Beatles, Past Masters, Vols. 1 & 2 (Capitol C21Z-90043 and C21Z-90044). Great-sounding classics up to the high standards of the other Beatles CD reissues; NoNoise was applied selectively to unidentified cuts.

The Doors, Live At The Hollywood Bowl (Elektra 60741-2). Reconstruction of severely damaged vintage 1968 live concert; excellent sound; no hint of processing.

The Grateful Dead, Europe '72 (Warner Bros. 2668-2, two CDs). Live recording substantially cleaned up, with some hiss audible at loud volumes.

George Harrison, All Things Must Pass (Capitol C22V-46688). Generally good, with several hissy pieces; mildly inconsistent sonic quality.

John Lennon, Plastic Ono Band (Capitol C21 Z-46770). Generally very good sound quality; occasional muddiness not due to NoNoise.

John Lennon, Rock 'n' Roll (Capitol C21 Z46707). Superb, consistently clean sound.

Liberace, The Best of Liberace (MCA MCAD-4060). Pop piano, with fairly steady low-level hiss noticeable in the solos.

Various, The Disney Collection, Vols. 1 & 2 (Walt Disney Records CD002 and CD003). Hard not to like; sonically uneven due to wide variety of difficult sources (including optical soundtracks) and processing done after editing, mixing, and equalization (the reverse of recommended procedure); still, clarity is remarkable.

===========================

This can be heard on Volume 3 of MCA Classics' The Segovia Collection (MCAD-42069), which features some of El Maestro's favorite works. To begin with, the classical guitar has such a fragile sound that it is already closely intertwined with hiss. When it plays forte, however, the hiss begins to recede. But as phrases conclude or the instrument turns dolce, the hiss surges or "pumps" back. You'll notice this elsewhere in the Philips' Legendary Classics series as well. I'm sure the effect was not intended by the producer, but you'll encounter it nonetheless.

There's a second related effect on the Segovia anthologies, however, which brings us back to the interesting issue of subjectivity. This involves fading to a pure, noise-free "black" which you'll hear between selections or movements. Just as the ear has adjusted to the prominent hiss, it disappears rapidly. This further serves to emphasize the background noise and would have been better handled by allowing the fade to taper more gradually, and perhaps by actually leaving a little noise in the holes.

All of the producers with whom I've spoken acknowledge that the subjective factor must be reckoned with, Independent producer Ed Michel, who has worked on many of the Bluebird reissues, says, "It's important to remember that NoNoise is a tool that's only as good as the way you use it." This point about subjectivity is reiterated by Michael Jarrett of EMI's Abbey Road Studios, who employed NoNoise on the CD of John Lennon's Plastic Ono Band (Capitol C21 Z-46770). "Once you start making aesthetic judgments," says Jarrett, "you're on dangerous ground. For example, did Lennon use dreadful-sounding effects at the beginning of Plastic because that's all the BBC library held at the time, or because he wanted the corny, funky sound? If you clean it up, are you distorting the intent?" Jarrett left the noise in the bell effect.

Although many consumers have come to expect no noise on their CDs, this may not always be practical or desirable. "NoNoise is the best available right now, and I really like the way it does its math," explains Grateful Dead producer Joe Gastwirt, who's also used NoNoise on numerous CBS, Jimi Hendrix, and other projects. "But if it comes down to hiss versus high end, sacrificing the highs for less noise isn't worth it." "We take a conservative approach," asserts Gene Wooley, Vice President of Recording and Quality Assurance at MCA. "The integrity of the music is most important to us. NoNoise is less destructive to the audio than other methods, but even at its best, it only allows you to achieve a happy medium, one which requires subjective judgment."

Once subjectivity is admitted, other questions are implicitly raised.

Gastwirt points out a concern that inexperienced producers can inadvertently affect the original material seriously.

"Tastes change in terms of what frequencies are preferred," he says. "It's important that a producer be familiar with the values at the time the material was recorded. And he must know how changing one part will affect another.

Not all producers working on old masters know these things."

The issue of altering intent, of altering history, becomes most poignant if record companies think that de-noising and digitizing old masters means they no longer need to keep the originals around, which is often an expensive proposition. This concern has been raised by PolyGram catalog executive Bill Levensen, who is very cautious about the NoNoise process. (Only The Yardbirds demo tapes on his recent Eric Clapton retrospective, Crossroads, were de-noised by Sonic Solutions.) "What happens," he queries, "when a better technique comes along, and the original art is gone?"

In the final analysis, it's important to keep in mind that NoNoise is a process that's intrinsically archeological in nature. That is, it's not intended, 'like Dolby noise reduction, to decrease noise buildup by encoding the music signal during recording and decoding it in playback. Rather, it is being used to post-process material which is already very noisy and often in a deteriorated condition. If the sources weren't imperfect to begin with, NoNoise would have no raison d'étre. Therefore, any judgments must be tempered by a comparative perspective. In many cases, employing NoNoise is the audiophilic equivalent of dealing with the ravages of time and acid rain on cultural artifacts like the Parthenon.

With the Philips license in Europe and Japan and the new Mac-II version for in-house use (which also includes digital editing, EQ, and mixing facilities), NoNoise has the inside track on becoming the standard processing methodology. Producers need to be sensitive to its limitations and to the philosophical implications of using such a tool. Audiophiles should evaluate the resulting product with a realistic understanding of what's going on.

But please, record companies, hold on to those original masters! The war against nasty noises isn't over yet!

(adapted from Audio magazine, Mar. 1989)

Also see:

Cedar Noise Reduction-- Getting the Grain Out (Mar. 1992)

= = = =

Prev. | Next

Top of Page    Home

Updated: Sunday, 2018-07-22 9:06 PST