Reply to thread

Message: [QUOTE="Bombastus, post: 91101, member: 2178"] CEDAR positions Voxis for conferences, but the plugin isolates human voices while an automixer controls how multiple microphone channels contribute. One cleans the audio that passes. The other decides how much of each microphone reaches the room, recording, or remote participants. Voxis recognizes speech and vocals against unwanted noise and excessive reverb. CEDAR does not advertise speaker enrollment, voiceprints, or person-specific selection, so it should not choose the chairperson over an interruption. Two human voices remain a microphone-placement and channel-control problem. That makes [B][URL='https://goldmidi.com/community/threads/cedar-audio-brings-voxis-voice-isolator-to-live-sound.76468/']conference voice isolation with Voxis[/URL][/B] a cleanup layer rather than channel-selection logic. A panel still needs automatic microphone mixing, sensible routing, and individual mics that capture each participant cleanly. Otherwise, every open channel keeps adding room sound even after one mixed output receives isolation. [HEADING=2]Automixing controls microphone contribution first[/HEADING] A gating automixer opens or raises microphones when people speak and attenuates them when they stop. A gain-sharing automixer distributes a consistent amount of system gain among channels according to their activity. Both reduce the buildup of unused microphones without asking an operator to chase every short exchange. Neither method removes HVAC noise from an active microphone while a panelist is speaking. Once that channel receives gain, its fan noise, keyboard clatter, and reverberant room sound travel with the voice. Voxis can work on that combined signal, which is why the two processes complement rather than replace each other. Gain sharing usually sounds smoother during overlapping conversation because it changes channel levels continuously. Gating can provide firmer separation, but thresholds and hold times need careful commissioning so opening words are not clipped. The right choice depends on whether the room prioritizes natural discussion, strict microphone control, or local voice lift. Audience questions need their own microphone path rather than faith in neural separation. A close handheld microphone gives the question useful direct sound, while the automixer can attenuate idle panel channels. Voxis may clean the active audience feed, but it cannot decide whether an off-mic audience voice has permission to enter the event. The resulting [B]controlled multi-microphone speech mix[/B] gives isolation a better input. Fewer unnecessary room feeds mean less diffuse noise for Voxis to interpret, while an active channel still receives detailed cleanup. This division also leaves the operator able to mute, prioritize, or record speakers independently. [HEADING=2]AEC and mix-minus protect the remote conversation[/HEADING] Acoustic echo cancellation uses a reference of the far-end audio played through the room loudspeakers. It compares that reference with what local microphones capture and removes the returning loudspeaker signal from the conference send. Voxis has no equivalent far-end reference and cannot perform that routing-aware calculation. Mix-minus solves another problem by withholding a participant’s own signal from the feed returned to that participant or nearby loudspeaker zone. The remote caller hears everyone except the caller’s delayed return, while local reinforcement can omit a microphone from the speakers closest to it. Voice isolation cannot create those destination-specific feeds. Keep the AEC reference tied to the signal that actually feeds the room loudspeakers. Voxis adds less than 10 milliseconds of stated processing latency, but inserting extra delay into only one relevant path can still change echo-canceller performance. Recheck full-duplex conversation after placement rather than judging the local microphone alone. Many conference processors already include steady-state noise reduction inside or near their AEC path. That processing may handle air-conditioning or projector fans before Voxis is added. Compare the built-in reduction, Voxis, and the combination because stacked cleanup can damage consonants or produce changing voice texture. [HEADING=2]Instance placement changes the conference result[/HEADING] Per-channel Voxis instances allow different low, mid, and high settings for a noisy lectern, ceiling array, and audience handheld. That flexibility costs more host processing and creates more paths whose latency and recall must remain predictable. It also lets each microphone reach the automixer with a differently processed noise floor, which can affect activity detection. One Voxis instance after the automixer uses fewer resources and receives the exact conference mix sent onward. The compromise is shared processing for microphones with different distances, rooms, and noise spectra. A setting that suits a close lectern may overwork a distant ceiling microphone when the active speaker changes. Conference DSP guidance commonly places noise reduction before automatic gain control or compression so later gain does not lift the untreated noise floor. External plugin placement must still respect the chosen AEC architecture and the host’s latency compensation. A tidy diagram is not proof that the reference, microphone, and transmit paths remain aligned. Commission the complete room with quiet and loud talkers, overlapping panel speech, an audience question, far-end double talk, and the HVAC running. Watch automixer activity while listening for clipped openings, residual echo, and isolation artifacts on the remote feed. If the caller hears room loudspeaker echo, correct the AEC reference or mix-minus routing instead of turning Voxis harder. [/QUOTE]

Name