Cochlear Implant Atlas
CI Atlas · From Sound to Stimulation · Module 08

8Peak-picking — SPEAK & ACE

CIS drives every channel every cycle. The strategies most widely used today take a different tack: analyse the sound into many bands, but in each cycle stimulate only the few bands carrying the most energy — the spectral peaks. This 'n-of-m' approach concentrates the implant's limited stimulation where the information is, and by leaving the quiet channels silent it sidesteps some of the channel interaction that plagues full-channel stimulation. SPEAK pioneered the idea; its faster descendant ACE became the default strategy in Nucleus devices and one of the most-used coding strategies in the world. This module explains how peak-picking works, why it helps, and what the clinician is actually choosing when they set the number of maxima and the stimulation rate.

TStimulate the peaks

An n-of-m strategy analyses sound into n frequency bands but, each stimulation cycle, selects and stimulates only the m bands with the highest energy — the spectral peaks— leaving the rest silent that cycle. As the sound changes, which bands are “in” changes with it, tracking the moving peaks of speech.

Pick the peaks — stimulate only the m busiest of n bands each cycle (ACE, SPEAK)

n = 12 analysis bands · ● = stimulated this cycle

Rather than drive every channel, an n-of-m strategy selects the m most energetic bands — the spectral peaks where the information is — and stimulates only those each cycle. SPEAK and its faster successor ACE (the default in Nucleus devices) work this way: it concentrates the limited stimulation budget on what matters and sidesteps some channel interaction. Choosing m trades spectral detail against stimulation rate. Schematic.

CSPEAK and ACE

SPEAK (Spectral Peak) introduced peak-picking with a relatively low, adaptive rate. ACE (Advanced Combination Encoder) kept the n-of-m principle but ran at a much higher stimulation rate, combining SPEAK's spectral selection with CIS-like temporal sampling. ACE became the default strategy in Cochlear's Nucleus devices and is among the most widely used strategies worldwide; MED-EL and Advanced Bionics offer their own CIS- and n-of-m-derived equivalents.[2015]

CWhy pick peaks

Peak-picking helps for two reasons. It puts the limited stimulation budget on the channels that matter — spectral peaks carry most of the speech information — and by not driving every channel at once it reduces channel interaction (Module 7). The silent channels are mostly carrying noise or low-energy detail the listener would not resolve anyway, so little is lost and interaction is gained back.

The two dials behind every strategy — temporal rate vs spectral maxima

maxima (spectral)stimulation rate (temporal) →SPEAKACECIS

High rate AND many maxima (~8–12) — combines SPEAK's spectral selection with CIS-like temporal sampling; the common default in Nucleus devices.

No corner is simply “best”: more maxima means richer spectrum but more channel interaction; a higher rate samples the envelope more finely but spreads the stimulation budget thinner. The right balance varies between recipients — which is why programming is individualised. Schematic positioning.

CThe knobs: m and rate

Two parameters define the trade-off. The number of maxima m sets how much spectral detail is delivered each cycle — more maxima means richer spectrum but more interaction; the stimulation rate sets how finely the temporal envelope is sampled. Higher rates and more maxima are not always better; the best settings vary between recipients, which is part of why programming is individualised rather than fixed.

The peaks move — the chosen channels track the formants frame by frame

chtime →blue = stimulated this frame

Peak-picking is not a fixed choice of channels — the spectral peaks of speech move as formants glide, so the set of stimulated electrodes is re-chosen every frame, chasing the energy. With a small m the selection hugs the formant tracks tightly; with a large m it fills in, approaching full CIS. Watching the blue cells follow the diagonal sweeps is peak-picking in action. Schematic.

Case 8.8 · Why not drive every channel?
A recipient programmed with ACE asks why his processor stimulates only some of the analysis bands each cycle rather than all of them, worrying that information is being thrown away.

What is the rationale for peak-picking?

Self-assessment — Module 82 questions
Question 1 · Trainee

How does an n-of-m strategy such as ACE work?

Question 2 · Clinician

Why does peak-picking help despite leaving some channels silent each cycle?

Tracked locally in your browser — see /progress for the dashboard.