The latest version of BIAS's popular restoration software aims to remove even more hassle along with your unwanted noise...
BIAS's SoundSoap Pro is one of the most popular affordable audio‑restoration packages on the market, and is capable of tackling a wide range of tasks including vinyl and tape transfer, music production and mastering, and cleaning up film soundtracks. The new version 2 adds some major new features, the most significant of which is an 'adaptive' broadband noise‑reduction mode that works in the absence of a noise fingerprint or where the noise floor is varying. There's a cost‑effective upgrade path for existing SoundSoap Pro customers, and SoundSoap Pro 2 is also now included in the Peak Pro XT 6 bundle. SoundSoap Pro 2 is cross‑platform, supporting Mac OS and Windows hosts running VST, Audio Units, RTAS or AudioSuite plug‑in formats, and now works with Pro Tools 8.
Whereas some companies prefer to provide a separate plug‑in for each type of restoration process, SoundSoap Pro 2 integrates them into one plug‑in, using buttons to select which screen is displayed for making adjustments. The processes are arranged from left to right in the order in which they are most likely to be required: Hum & Rumble filters, Click & Crackle elimination, Broadband noise and finally Noise Gate. The crackle, hum and gate sections are pretty much the same as in the original SoundSoap Pro, though the user interface has been streamlined, with new tooltips and display features, but there's a new 'ANR' adaptive mode in the broadband section. As we've covered SoundSoap Pro in its original incarnation, and the full review is available to read for free on our web site (/sos/feb05/articles/soundsoap.htm), I'll concentrate mainly on the new features in this review.
While some restoration software, such as the excellent Sonnox Suite, encourages forensic adjustment to optimise performance, SoundSoap Pro has always aimed to make setup as simple as possible, though in my experience some user adjustment is still necessary to get the best results from it. For example, SoundSoap Pro uses a 512‑band filter to split the audio spectrum for processing, but as far as adjustment is concerned, there are just 12 double faders, each of which offers separate adjustment of the threshold and noise-reduction amount in its part of the audio spectrum. The threshold and noise-reduction faders can either be left locked, so that moving one moves all together, or they can be unlocked to allow individual adjustment.
Whereas a few seconds of exposed noise were needed for the most effective broadband noise reduction in the original SoundSoap Pro, it is now possible to choose between the traditional fingerprint method, whereby a section of 'noise only' is analysed, or the new ANR option, which uses algorithmic methods to decide what is signal and what is noise spectrum. ANR is clearly useful for audio that has been previously trimmed to remove any 'noisy silence' at the start and end of the file, but it is also effective where the noise gradually changes in character throughout a recording. In all modes, the user can solo the part of the signal identified as noise, using the Noise Only button, so it is possible to check that no significant part of the wanted signal is being removed.
As well as the automatic ANR adaptive noise reduction, there's also an adaptive noise Extraction mode that requires some adjustment by the user, the purpose of which is to extract a noise profile from an audio track with no exposed noise‑only sections. Another new feature is the Preserve Attacks tick-box that makes the noise-removal process kinder to transients. As I understand it, this applies less noise reduction during transients, to minimise apparent treble loss.
The Learn button, normally used to create a profile from a noise‑only segment, changes to 'Extract' when the Adaptive button is on, and it is this that gives access to the Extract mode of operation, where an average noise fingerprint or profile is extracted from a section of audio that contains both noise and signal. This mode is best suited to audio where no noise‑only section is available and where the noise character and level remains reasonably constant throughout the file.
Extraction can take up to 30 seconds or so (the exact time being decided by the software) and the noise profile created can be improved by adjusting the Speed control, which essentially adjusts the rate at which the software averages out changes in the noise character while creating its profile. If the noise is reasonably constant, a slower Speed setting produces the best results. Once the noise profile has been extracted, it is used just as in earlier versions where you had a noise‑only section to analyse, and the user adjustments are basically the same. The user can also manually choose the section to be analysed in Adaptive mode, using the Timed Extraction button (the clock icon immediately to the right of the Extract button) to start and stop the analysis manually, rather than letting the software decide. This may be appropriate where the section of audio is shorter than the automatic analysis time or where the user feels it beneficial to average the noise over a longer period of time.
Fully automatic Adaptive mode, accessed by selecting Adaptive but not subsequently pressing Extract, attempts to create a changing noise profile on the fly. During fully automatic noise removal, the Speed knob determines how quickly the process follows a changing noise spectrum. Again, the slower you can set this, the fewer artifacts are likely to be generated. During my tests, I needed to set the control past its midway position to avoid chirps and warbles. The individual threshold sliders are not available in Automatic mode, as the software is constantly adjusting them but there is an Offset knob that can be used to move the overall noise profile up or down by a few dB. There are also attack and release controls that set the time constants for the noise-reduction process when the signal crosses the threshold in each frequency band. A further pair of controls called Attack Tilt changes the attack and release time between the high and low frequencies, so that the time constants are shorter at high frequencies than at low frequencies.
BIAS SoundSoap Pro 2's crackle and hum sections work as before and are reasonably effective in improving problems such as vinyl surface noise and mains hum. The gate, too, is very straightforward, and would normally be used to silence sections between wanted parts of the audio file where no signal other than residual noise is present.
As a fingerprint noise‑removal solution, the original SoundSoap has always provided an effective middle ground between cheaper plug‑ins that don't produce particularly good results and the big‑bucks solutions that may also take more experience to use effectively. As long as you could find a noise‑only section in the audio file and you didn't try to attenuate the noise by more than was strictly necessary, the results could be very good indeed. The new modes make the software more versatile, as it is no longer imperative to find a section of exposed noise to analyse, but if you can find a few seconds of exposed noise, and the noise is essentially consistent, this mode still produces the most natural results.
The new Extract mode works in essentially the same way as the original noise fingerprint mode, except that a viable noise profile can be created where no exposed section of noise is available. As long as the noise is consistent and the speed setting is set past midway, this works almost as well as having a noise‑only section to work with, though if you have the patience to adjust the individual threshold and noise-reduction amount sliders, you can often obtain a cleaner‑sounding result than simply relying on the default settings. For example, noise is often the most intrusive in the upper-mid range, whereas above 7kHz it becomes less noticeable, so if you set less gain reduction both at lower frequencies and above 5kHz or so, there tends to be less 'burbling' and less apparent loss of high end.
The fully automatic Adaptive mode works adequately well as long as you don't set the Speed knob too fast, as this can introduce audible artifacts. As ever, if you try to attenuate the noise too much, the dreaded 'chirpy and bubbly' artifacts start to appear, though tweaking the gain‑reduction amounts in the various bands can really help minimise this. I found that adjusting the individual gain‑reduction faders while listening to the change in noise was quite effective in reducing artifacts, as you can then apply a little more gain reduction to the bands that need it and reduce it in those that don't. From watching the noise profile as outlined by the threshold faders, I found that SoundSoap Pro 2 seems to assume there is more noise than there is when the wanted signal level falls. If you can hear the noise removal working in Auto mode, therefore, it is worth also adjusting the Offset control to bring the threshold down a bit, so you can set the subjective best balance between minimising the noise and changing the character of the audio. Remember that in this mode you can't access the individual threshold faders, as they are always moving under software control.
To test the Adaptive mode's ability to respond to a changing noise floor, I created a multitrack project including normalised recordings of guitar amplifier hiss, computer fan and drive noise, and circuit hiss. Using mix automation, I then arranged for these to change in balance and level and added them into my clean audio at a level of around ‑40dB. Letting Adaptive mode run entirely automatically, the start of the audio sounded distinctly watery as the algorithm struggled to generate a noise profile, but after a few seconds, it settled down and seemed to work pretty well. To minimise this problem in real life, it would be practical to copy 30 seconds or so of the audio to the front of the file to allow SoundSoap Pro 2 to stabilise before it reached the wanted section of audio. This could then be trimmed off again after denoising.
However, Adaptive mode didn't respond well to fast changes in noise, as setting the Speed control lower than halfway made the audio sound very watery, and longer settings took 10 or more seconds to adapt to the new noise profile. In other words, don't expect miracles where the noise level and/or spectrum changes radically. Even on slower speed settings I found the audio sounded a little artifact‑ridden unless I adjusted the noise‑reduction faders and set each for as little gain reduction as possible, so unless the noise pollution is lower than around ‑50dB, don't expect to achieve perfect results just by pressing 'Adaptive' and sitting back.
I don't think BIAS would ever claim that SoundSoap was the most effective restoration plug‑in on the market for broadband noise reduction, and I have had better results (with fewer audible artifacts and less high‑end loss) with TC's Denoiser for PowerCore and Sonnox Restoration Suite. However, not all competing plug‑ins can work without an exposed section of noise to analyse. One that can is Waves' Z‑Noise, and with careful adjustment I found that I could get just as good a result using the Adaptive mode in SoundSoap Pro 2 as I could in Z‑Noise. If you don't ask too much of it, SoundSoap Pro 2 can bring about significant improvements in all kinds of noise contamination without demanding too much in the way of expertise from the user. As with much of the competition, it struggles to keep the audio sounding natural where the noise contamination is high in level, but for modest amounts of circuit or tape hiss, or noise from air conditioning, computers and cameras, it can work effectively, and it really doesn't take long to learn how to tweak it for the best results.
Although SoundSoap Pro 2 can't be considered a magic bullet to fix all noise contamination issues, its automatic Adaptive and noise Extraction modes make it more useful than ever, as you often get audio to deal with that has no exposed noise‑only sections. Even the best noise‑reduction software will compromise audio quality if pushed too hard, so SoundSoap Pro 2 always needs to be used as sparingly as possible, but given its affordable price, ease of use and flexibility, it still represents a practical and pragmatic solution for many routine audio clean‑up problems involving moderate levels of hiss, hum, buzz and crackle.
Low‑ to medium‑cost alternatives include TC's denoising package for the PowerCore platform, Waves Denoise, and the slightly more costly and more sophisticated Sonnox Restoration Suite.
The most common type of broadband noise reduction works by splitting the audio into a large number of frequency bands. Each of these is then fed through an expander circuit, allowing signals falling below a certain level to be muted or attenuated. It would be impractical to set hundreds of separate thresholds manually, so instead a noise 'fingerprint' or profile is taken by analysing a short section of audio containing only noise, usually from before the start of a piece of audio or during a pause. The amplitude of the noise within each frequency band is measured, and this is used to set the threshold value for that band. If no noise‑only section is available, it is possible to use a generic noise-spectrum profile based on, for example, tape hiss or analogue circuit hiss, adjusting the overall threshold manually until the best results are achieved. There's usually a facility to audition just the sound being removed as you adjust the overall threshold, so you can hear when parts of the wanted signal are being affected. This was how the original SoundSoap worked.
The next stage of refinement is to allow the software to analyse a section of audio and to use an algorithm to try to estimate what is wanted signal and what is noise. Real audio is constantly changing but noise tends to be either reasonably constant or slow‑changing, so by taking a long enough sample, the algorithm can make a reasonable prediction as to what is noise and what is wanted audio. The capture time will vary depending on the material, so in some systems the software decides how much audio it needs to 'see' to get a reliable result but gives the operator the option to change this if necessary. The noise data is then used to create a noise profile based on the average noise amplitude and spectrum over the sampling period. The clear benefit of this method is that you can work on audio that has been trimmed to remove any noise‑only sections. It can also be helpful on occasions to choose a section of audio where the noise is most exposed when creating a noise profile. However, it is only really effective when the noise character is fairly consistent.
The Holy Grail is an algorithm that can figure out what is noise, even when the noise character is changing, and then adjust the expander thresholds to follow the changing noise profile in real time. For this to work effectively, the noise character still needs to change more slowly than the audio, otherwise the software has no way to tell what is wanted audio and what is noise, so rapid changes in noise character, such as might occur at an edit point, may be exposed for a short time before the software has time to respond. On SoundSoap Pro, you can adjust how quickly the algorithm follows changes in the noise spectrum, but the compromise is that the faster you set the response time, the shorter the period of analysis and, as a consequence, the algorithm will be less precise in discriminating between noise and wanted signal. In theory, the more rapidly the wanted signal is changing, either in frequency content or amplitude, the faster the noise tracking can operate without tripping up.
In all cases, manual adjustment of the overall threshold level may be required to get the best balance between noise removal and processing artifacts, and this is best done by auditioning the removed part of the signal. It is also important to attenuate the noise only as much as is necessary to avoid noticeable processing artifacts such as loss of treble or a watery, 'chirping' character. In some cases, making two passes over the same audio with a relatively low level of attenuation will produce more natural‑sounding results than a single pass with twice the level of attenuation.