8Front-End Processing: IDR, Sensitivity and Noise
Before any coding strategy fires a single channel, the sound processor's front end has already made decisive choices: how much of the acoustic world to let through, how loud to make it, and how to favour the talker over the room. Input dynamic range, microphone sensitivity and automatic gain control together squeeze a 100-plus-decibel acoustic landscape into a narrow electrical window of perhaps 40 to 60 dB. Layered on top are directional microphones, noise-reduction algorithms and scene classifiers that decide, moment to moment, what counts as signal.
TInput dynamic range: the acoustic window
The input dynamic range is the span of acoustic input levels mapped into the recipient's electrical dynamic range; the default on contemporary devices is roughly 40 to 60 dB. The lower edge, usually set between about 20 and 35 dB SPL, defines the softest input mapped near electrical threshold, while the upper edge, set between roughly 65 and 90 dB SPL by manufacturer, is mapped near the maximum level. Inputs below the floor fall under electrical threshold and are inaudible; inputs above the ceiling are subjected to high-ratio compression so they do not exceed comfortable loudness. A wider input dynamic range improves audibility of soft speech and environmental sound but raises the audibility of background noise; a narrower one cleans up noise at the cost of soft-speech access. Cochlear additionally uses an instantaneous input dynamic range, typically about 40 dB, that captures the peak-to-valley range of ongoing speech without compression, with the sensitivity control sliding this window up or down across a much wider total input range.[2020][2014]
TSensitivity, AGC and front-end compression
Microphone sensitivity sets the gain applied to the input before frequency analysis, so its effect is broadly equal across the frequency range; raising it improves audibility of soft sounds, lowering it suppresses low-level noise. On a contemporary Nucleus processor the sensitivity control runs on a numbered scale with a default near 12; reducing it raises the input level needed to engage the high-level compressor and lifts the input-dynamic-range floor. All three manufacturers use automatic gain control as the front-end compressor that funnels a wide acoustic range into the narrow electrical dynamic range, but the gain functions and time constants differ markedly between systems. Compression behaves as an automatic gain control whose responsiveness is governed by attack and release time constants; fast-acting compression protects against transients while slow-acting compression preserves the speech envelope. Channel gain offers a frequency-specific adjustment layered on top of the global sensitivity and AGC, letting the clinician boost or trim individual channels, for example to add high-frequency audibility, independently of the overall input mapping.[2020][2014]
CDirectional microphones and scene classification
Most processors carry two omnidirectional microphones that combine into a dual-microphone beamformer; an omnidirectional mic at the top of the ear actually gives a negative directivity index that can degrade speech-in-noise. The Nucleus dual-mic system offers graded modes, from a standard setting that mimics the natural unaided directivity, through a fixed forward beam with strong rear attenuation, to an adaptive mode that steers the null toward the loudest noise, and an automatic classifier that picks the mode for the detected scene. Advanced Bionics offers an adaptive monaural beamformer, a bilateral four-microphone beam that outperforms the monaural beam in noise, and a fixed beam for situations like a car passenger. Placing the microphone at the ear-canal opening recovers the pinna effect and improves speech-in-noise compared with a top-of-pinna omnidirectional mic. Adaptive noise-reduction algorithms make channel-by-channel decisions and attenuate channels judged to be mostly noise, with selectable strength of up to about 6 dB (low), 12 dB (medium) or 18 dB (high).[2020][2014][2014]
CStreaming inputs and multiple environmental programs
The sensitivity parameter on newer processors governs the gain of both the microphone and the direct audio input, so streamed audio shares the same front-end mapping into the electrical dynamic range. Modern processors stream via 2.4 GHz wireless protocols and integrate a telecoil, plus remote-microphone and FM/DM accessories that move the microphone to the talker and can substantially improve speech recognition over distance and noise. Wireless remote-microphone and audio-streaming accessories improve speech recognition relative to the on-ear microphone alone, because they bypass the room acoustics between talker and listener. Because the optimal front-end settings differ by environment, recipients are commonly given multiple programs, for example an omnidirectional wider-range everyday map, a directional plus noise-reduction map for restaurants, and a streaming or telecoil program, selectable by button or app. When noise reduction is enabled the clinician must re-confirm overall loudness, since attenuating noisy channels lowers perceived level, and the algorithm needs a short analysis window before it begins attenuating unfavourable channels.[2020][2014][2014]
Which front-end change is the best first step for his restaurant difficulty?
Lowering microphone sensitivity on a cochlear implant processor primarily does what?
A typical medium setting of an adaptive channel-by-channel noise-reduction algorithm applies up to how much attenuation?