AutoMix sidechain compression for cleaner live voice
AutoMix’s Sidechain Compression Integration gives live producers a focused tool for controlling sibilance without flattening the character of speech. By using a separate detector to trigger dynamic gain reduction, it can soften harsh /s/, /z/ and /sh/ sounds while leaving the body of the voice intact.
That precision matters when a presenter moves between prepared copy, spontaneous talk and caller audio. Instead of applying a permanent high-frequency cut, the processor responds only to the consonants that need attention, preserving natural phrasing, intelligibility and consistent loudness across a live show.
Why live voices need dedicated de-essing
Live radio microphones capture a wide range of accents, room reflections and performance levels. Peaks created by plosive consonants can be distracting, while excessive sibilance may sound harsh in headphones, on a studio monitor or through a car radio. The same peak can also force a level rider to pull down the entire voice.
A static equaliser treats every consonant as if it were the same problem. Sidechain de-essing is more selective: the detector identifies high-frequency energy, and a compressor reduces gain only while that energy is present. The voice then returns to its original level as the fricative ends.
How the sidechain detector shapes gain
AutoMix’s integration provides a trigger path for a compatible compressor in the live signal chain. The sidechain path acts as the key, while the assigned voice path is the signal being controlled. For de-essing, a filtered version of the voice is fed into the compressor’s key input; once it crosses the threshold, gain reduction follows the peak and releases as the consonant decays.
Threshold sets where the detector begins working, ratio determines how strongly it responds, and range limits the maximum reduction. Attack and release control how quickly the compressor catches and recovers. Short attack preserves the consonant’s definition, while a controlled release avoids audible pumping between words. Make-up gain can then restore the perceived level.
Building a practical voice chain
A reliable chain keeps the detector reference aligned with the voice it will process. The microphone preamp and conversion stage should operate linearly, the de-esser should sit before aggressive saturation, and the programme feed should be monitored before limiting. On Lawo Ruby consoles, operators retain direct control of source selection, talkback, monitor paths and scene recalls.
Several processing choices differ in what they actually respond to. The useful comparison is between dynamic detection, continuous tone shaping and operator control.
| Processing approach | What triggers action | Result on the voice | Main limitation |
|---|---|---|---|
| Sidechain de-essing workflow | Filtered high-frequency detector | Gain reduction follows sibilant peaks | Requires representative tuning |
| Broadband leveller | Overall signal level | Evens loud and soft speech | Can treat sibilance as general loudness |
| Static high-frequency cut | Fixed setting | Reduces sibilance consistently | Alters every word in the same way |
| Manual gain riding | Operator judgement | Preserves full dynamics under close supervision | Inconsistent during busy live periods |
For live speech, dynamic sidechain control usually provides the best balance: it reacts to the problem rather than imposing a permanent tonal change. A leveller can precede it for gross level control, while careful gain riding remains useful for expressive delivery.
Tuning for real Australian speech
Calibration should use the actual mix, not an isolated spoken sentence. Australian English is varied, and a host in Sydney, a caller in Brisbane and a panelist in Melbourne can produce quite different consonant peaks. Record representative lines, including plosives, and set the detector so ordinary speech never causes visible gain movement.
That approach avoids false action while still catching genuine sibilance. The same distinction between useful detection and unnecessary rejection appears in optical systems, where reducing false rejects depends on thresholds matched to the real signal rather than an extreme test condition.
Listen through the complete monitoring chain, including headphones, talkback and the station’s broadcast processor. A setting that sounds controlled on the desk can still become harsh after encoding, so final checks should include the on-air feed and multiple playback systems.
Remote workflows across Australia
National workflows add another layer. A network can link studios in Sydney and Melbourne with contributors in Brisbane, Adelaide, Perth or Darwin, and each path may arrive at a different level. Power Core audio nodes and RƎLAY virtual radio tools can support the wider distributed path, but AutoMix still needs deliberate detector assignments because codecs and local chains alter the sound.
Remote interviews also need isolation from local microphone processing. The principles behind clean phone-interview audio use N-1 bus management on Ruby so an interviewer’s microphone does not feed back into a guest’s return path. De-essing should likewise reference only the intended voice, never the full programme bus.
For Australian broadcasters, including commercial, community, ABC and SBS operations, predictable settings simplify handover and reduce dependence on one engineer’s ears. Equipment may be supplied through broadcast integrators, but approved scene values should remain under the station’s control and align with ACMA requirements and the network’s loudness target.
A controlled deployment
A controlled rollout begins in rehearsal with one host and one remote contributor. Capture gain reduction alongside the audio, compare the processed voice with an untreated reference, and document any rapid recovery or loss of consonant clarity. Re-test after changes to microphones, preamps, codecs and the station processor.
The next concrete step is to run one supervised session using representative Australian voices, record the chosen threshold, ratio, range, attack and release, and save those values as the approved AutoMix de-essing scene. That settings record becomes the handover reference for every later shift.
