By Jennifer Walden
Steinberg’s SpectraLayers Pro 11 Advanced Spectral Audio Editor isn’t the typical sound restoration tool. This latest version includes AI-assisted processes that help editors separate mixed tracks and repair recordings with precision and ease.
Yes, that’s right: separate mixed tracks! Their “Umix” modules offer un-mixing options for dialogue and music, including: Chorus, Components, Crowd Noise, Drums, Levels, Mid/Side, Multichannel Content, Multiple Voices, Noisy Speech, Song, and Transcription. After unmixing a sound file, SpectraLayers Pro 11 creates separate layers for each element (e.g., voice, noise, instrument, etc.) so that each layer can be processed and refined individually. The editor can rename each layer and change the layer color for easy identification.
Emmy-winning sound designer/sound effects editor/re-recording mixer James David Redding III, MPSE — known for his work on Netflix’s “The Queen’s Gambit,” director Todd Haynes’ film “Carol,” and the FX series “The Patient” — described SpectraLayers Pro 11 as “iZotope RX on steroids.” He first discovered SpectraLayers while working on a podcast series that had been recorded as mixed tracks. The overlapping voices made it impossible to make changes, like adjusting the timing to make room for effects or music, or rebalancing the voices.
“I used the Unmix Multiple Voices module, and it worked! While it didn’t get everything 100% right, especially when the overlapping voices sounded similar, there’s the option to cut from one layer and paste to another layer, so I could manually fix it,” he said. “The music stem separation is really good, too, especially compared to RX’s Music Rebalance feature or RipX. Look, no one tool is a silver bullet, but if I had to recommend one to someone who’s just starting out, I’m going to tell them to get SpectraLayers.”
UNMIXING MULTIPLE VOICES
The Unmix Multiple Voices module allows the editor to unmix an unlimited number of voices, but only up to three simultaneous overlaps. The editor must create a voice profile for each person in the mixed track by choosing between two available options in the module’s pop-up window: Register Voice or Register Multiple Voices.
Using Register Voice, the editor must highlight a place in the timeline when only one person is speaking, and then click Register Voice to create a voice profile. The profile will then be listed in the pop-up window as Voice 1 and can be renamed. This process must be done for each person speaking in the recording to give the AI algorithm a clean reference for each voice.
The other option is Register Multiple Voices. The editor enters the number of voices, and the algorithm determines and separates each voice. This is useful for quickly registering many voice profiles if the voices are clearly distinct from each other or if some voices are only heard in overlap with other voice profiles. Steinberg recommends first running the Unmix Noisy Speech or Voice DeNoise processing if the recording contains background noise.
After running the Unmix Multiple Voices module, each voice will be on a separate layer, accessible in the Layers section of the right-side panel. By soloing a layer, the editor can check for any errors. As Redding pointed out, sometimes the wrong voice will be on the wrong layer, but a simple cut and paste in the timeline can move it to the correct layer.
DIFFERENT MODULES FOR DEEPER CLEANING
Additional processing may be needed on each layer to clean up any noise or to remove residual frequencies. SpectraLayers Pro 11 offers several processing options. In the Modules menu, located in the rightside panel, the editor can choose to sort the modules by function: Unmix, Utilities, Restoration, or view All Modules. Sorting by Restoration will show modules such as DeBleed, DeClip, DeEss, DeHum, DeNoise, DeReverb, Voice DeClip, Voice DeNoise, and Voice DeReverb. A few helpful processes for this case of separating overlapping voices are Voice DeNoise and DeBleed.
Using the Voice DeNoise module on a selected layer, the editor can choose what type of background noise to attenuate. In the module’s pop-up window, there are Preset choices for: Default, Music, Noise (moderate), Noise (strong), and Other Voice, and the same choices are available in the Background drop-down menu. Trying different combinations of Preset and Background options, such as Preset: Noise (moderate) and Background: Other Voice and vice versa, will produce different results. By changing the option of Reduce Background to Reduce Signal, and hitting the Preview button, the editor can hear what the processing will remove. Adjusting the Reduction Ratio slider allows the editor to further control the amount of attenuation. Note that when using the Other Voice option (for two overlapping voices), the module will reduce the voice with the lowest RMS (i.e., the lowest perceived loudness).
The DeBleed module is an audio repair process that can help remove residual frequencies from overlapping sources. Although meant for attenuating bleed between layers of different instruments recorded simultaneously with multiple mics, it can also be used for voices if done with a light touch.
In the Layers menu, select one layer to DeBleed. In the DeBleed pop-up window, checkmark the source layer(s) to use as a reference for the DeBleed process. (This will likely be the original audio file imported into SpectraLayers, or the registered voice profile(s) the editor wants to remove from that layer.) From the Preset list, the editor can choose from Default, Light, Medium, or Strong, and further tweak the parameters by adjusting the Sensitivity and Reduction Ratio. There’s also the option to Reduce Bleed or Reduce Signal in the drop-down menu at the bottom; switching between the two and previewing the result will help the editor to dial in the processing.
For another recent project, Redding was asked to help save the dialogue in a hurricane scene. In the show, several characters are running around in a violent storm with strong winds and heavy rain. He said, “They shot with two boom mics, and there was a wind machine and rain machine running on-set. It was pandemonium, but the showrunners didn’t want to do a ton of ADR. Using SpectraLayers, I saved about 90% of the dialogue in the scene. The two lines I couldn’t save were when the principal actors went off-mic. I couldn’t clear it out because the lines just weren’t there. The fact that you can separate the recording into layers gives you so much flexibility. You can even use the background layer as production ambience — if there isn’t a generator or machine noise on it.”
Redding is also an Adjunct Professor at New York University and teaches a Sound Mixing Workshop. Using SpectraLayers Pro 11, he helped a student remove a generator sound from their production dialogue track. “There was no degradation of the recorded voices. The generator was just gone,” he said.
ADJUSTING RESOLUTION FOR IMPROVED RESTORATION
While working as a sound effects editor on the “Mr. & Mrs. Smith” (2024) series for Amazon Prime Video, Redding wanted to give a subway scene an authentic NYC transportation sound. “I needed that perfect subway door closing sound and, of course, the two-tone chime. Those sounds are so iconic. People will know if you have it wrong. I carry around a little Zoom M/S microphone for my phone, so I was able to record the subway sounds. But it wasn’t the best recording since I was in the crowded NYC subway. When I imported the recording into SpectraLayers, I could see the two pure tones of the subway chime stand out from this massive background of noise. I highlighted it with the Lasso tool, got rid of everything else, and there we go. It all comes down to being able to zoom in and being able to increase the resolution. Then you can see exactly what you want in the spectrogram. It’s funny because all these tools are so visual. SpectraLayers feels like Photoshop.”
In SpectraLayers Pro 11, the project display parameters can be quickly adjusted via the Display panel located in the rightside panel. Choosing “All Settings” in the drop-down menu opens an extensive list of options. The editor can set the minimum and maximum amplitudes shown in the spectral display, adjust the brightness curve, manually define the frequency range to be modified/processed, set the spectral display frequency range, choose the FFT window type (which has an impact on the spectral dynamic range), adjust the Resolution in sub-pixels for the FFT analysis, adjust the Refinement percentage (this will sharpen the spectral display so the editor can see the exact frequencies and timings more clearly), and adjust the FFT Size. The default mode of “Essential Settings” shows only Min/Max Amplitude and FFT Size. [FFT — Fast Fourier Transform — is a mathematical algorithm that decomposes a complex audio signal into its constituent frequencies.- ed.]
Adjusting the FFT Size allows the editor to bring different parts of the sound into focus as they edit by choosing between time and frequency accuracy. An FFT Size of 2048 or 4096 is good for most situations. A smaller FFT Size gives more details with time-centric events, such as transient sounds, but the frequencies will be blurrier. A larger FFT Size shows sharper frequency details, but the transients will be blurrier.
Redding said, “You can increase the resolution on the spectrogram for when you’re doing meticulous surgery on sounds. These display parameters aren’t hidden in some preference menu, so it’s easy to quickly change the resolution. If upping the resolution is using up too much computing power, then you can increase it when needed and then pull it right back down so you’re not choking your system out.”
There’s a 3D Perspective adjustment tool — located next to the Display panel — that allows the editor to change the perspective of the spectral display. Instead of seeing the spectral image in a flat, 2D view (that looks like a reading on a seismograph), this feature gives the editor a 360-degree-adjustable 3D view (that looks like an aerial shot of mountains) that’s based on spectral amplitudes. It’s an interesting and clarifying way of looking at the spectral image, and it makes editing the spectrogram much more precise.
VISUALLY EDITING SOUND
As Redding mentioned, spectrally editing sound feels very similar to editing visuals. They even share some of the same tools. In SpectraLayers Pro 11, the left-side panel contains a group of common editing tools, like the Lasso selection tool, Magic Wand selection tool, Selection Brush, Frequency Selection/Harmonics Selection tool, Transient Selection tool, Selection Fade, and Selection Sharpen. There’s the Eraser tool with many options for controlling what and how much to erase. The editor can choose to use a Round Brush or Square Brush, define the units used to set the brush shape proportions as either Screen Units or Spectral Units, define the brush Size in pixels, change the Aspect Ratio of the brush, define the Width of the brush as seconds, define the Width of the brush as hertz, set the Hardness of the brush (how sharp or blurry the edges are), choose to Limit Level of eraser attenuation and then manually set that limit in decibels. There’s even a Pressure Controls Strength icon that, when selected, will enable devices with pressure control to set the strength of the tool. So if an editor is using a stylus, the harder they press, the greater the attenuation.
The Amplifier tool increases the volume of the frequencies under the brush. The Transfer tool losslessly transfers content from the selected layer to a target layer. So instead of doing cut/paste from one voice layer to another, the editor can use a brush to “paint” out the voice from one layer and it will be transferred to another layer. The Clone Stamp tool copies content from an editor-defined source location to the current brush location. This is helpful for fixing missing consonant sounds, such as ‘t’ or ‘k,’ at the ends of words. The Frequency Repair tool can accurately repair a damaged frequency between a start point and an end point by using the frequency data of these two points to reconstruct what’s missing. The Frequency Pencil tool can be used to manually draw a continuous frequency. The Transient Pencil is used for drawing in a transient. There’s even a Noise Spray tool for “spraying” in noise.
Redding said, “I am a visual person. I grew up doing photography and using Photoshop. SpectraLayers gives you that extra tool set that you wouldn’t think of using for sound, like an eraser and a pencil.”
PLAYING NICE WITH OTHERS
With SpectraLayers Pro 11, the editor can import multiple audio files and have them layered in the same project. If the timing of each layer doesn’t perfectly line up in the project, the editor can note the start of one layer and use that as reference. Then, in the next layer, they can figure out how much time needs to be added to make it line up with the reference layer. In the Edit tab, they can choose “Insert Time,” set the Start point (where they want to add in time) and Length (how much time they want to add). This will shift the audio in that layer so it lines up with the reference layer. For instance, let’s say an interview between two people was recorded on separate devices. When importing both of those files into the same SpectraLayers project, the editor discovers that both voices start at 00:00:10.000. So when the two layers are played together, they overlap. The editor can “Insert Time” into the second layer, shifting the start of the voice so it begins after the first voice is done talking. Importing multiple layers is useful for editing music stems and effects layers, as well.
Editors can use third-party VST3 plugins inside SpectraLayers. For instance, if an editor prefers using a DeClick plugin from another company, and it’s available as a VST3, the editor can use that processing in SpectraLayers instead of the built-in DeClick processing.
If an editor prefers to work only within their DAW environment (and not deal with the import/export issue of using standalone software), they can use SpectraLayers as an ARA (Audio Random Access) plugin. This isn’t a watered-down version of the application with limited features. Instead, it looks (and sounds) exactly the same as the standalone application, with all the same tools, modules, and features. ARA 2 technology (designed to build bridges between audio editing applications) allows SpectraLayers to be launched inside a host DAW and acts as an editing feature. Redding said, “It’s great that you have this ability to do all these adjustments in the DAW like you would in the standalone SpectraLayers application.” Currently, the ARA SpectraLayers plugin is supported by: Steinberg Cubase 10.0.30 and higher, Steinberg Nuendo 10.0.30 and higher, Avid Pro Tools 2024.10 and higher, Cockos Reaper 5.98 and higher, Presonus Studio One 4.6
and higher, and more.
In Redding’s Pro Tools session template, he has a SpectraLayers worktrack, which has the ARA plugin already set up. He can simply drag the audio clip he wants to process onto the worktrack, edit it as needed, and then drag the edited clip back onto its session track. This will automatically apply the processing and generate a processed clip. “This is helpful if you’re sharing a session with an editor who doesn’t have SpectraLayers. This way, the processing is baked in. But if I want to keep the processing ‘live’ because I’m still tweaking it, I can simply make a track with the SpectraLayers ARA plugin, and as long as I’m staying on my system, I can keep it live in my session,” he said.
NO RISK, ALL REWARD
A full license for SpectraLayers Pro 11 is just $299, yet this lower price point doesn’t mean lower quality results compared to its competitors. SpectraLayers Pro 11 offers a massive selection of highly adjustable tools, uses new AI algorithms for improved speech and music repair, gives editors the ability to create processing chains, do batch processing, add in third-party VST3 effects, and it can be embedded into standard DAWs for a seamless workflow, making it an invaluable addition to a sound editor’s toolbox.
This article just scratched the surface of what SpectraLayers Pro 11 can do. Want to put it to the test against your go-to but don’t want to shell out hundreds of dollars? For owners of certain competing spectral audio restoration software, Steinberg’s Competitive Crossgrade discount drops the price to $199 for a full license. Still on the fence? Steinberg offers a free 30-day trial of the full version with no restrictions. The video tutorials on Steinberg’s YouTube channel are easy to follow, as is the user manual (for those who like to get into the weeds), so the learning curve isn’t steep. Redding concluded, “SpectraLayers is so powerful. It really shines in those Unmix modules. Having this level of control is incredible. I’m constantly telling people about this software.”
Jennifer Walden, who specializes in postproduction technology, is a regular CineMontage contributor.
