CI Atlas · Speech-Coding Strategies: The Complete Lineage · Module 01

1The Lineage at a Glance: Why Coding Strategies Keep Changing

Every cochlear-implant sound-coding strategy exists because the previous one hit a wall. This opening module lays out the genealogy as one branching family tree — single-channel to waveform to feature-extraction to n-of-m to fine-structure/current-steering to present synthesis to future — and establishes the two engineering axes (waveform vs envelope, all-channels vs peak-picked) that organise everything that follows.

FA four-decade family tree

The history of CI coding moves away from explicit speech-feature extraction toward a filter-bank / waveform approach, a transition dated to the early 1990s The genealogy runs single-channel (3M/House, 1970s) to multichannel waveform strategies (Compressed Analog, then CIS) to feature-extraction (F0/F2, F0/F1/F2, MPEAK) to hybrid/n-of-m (SPEAK, ACE) and the pure-spectral SMSP More than 120,000 patients had received cochlear implants by 2009, and the one-millionth implant milestone was celebrated in 2022 The modern era of cochlear implants is traced to Djourno and Eyries in France in 1957.[2008][1998]

~8Effective spectral channels ceiling for CI users (Friesen 2001) [2001]

20Channels at which normal-hearing listeners still improve [2001]

TThe motivating-limitation chain

Each strategy is best understood by the limitation that killed its predecessor: single-channel ignores cochlear place coding; CA's simultaneous stimulation causes channel interaction; CIS is fixed-channel envelope-only; formant trackers fail in noise; later strategies chase fine structure and spectral resolution A key overall finding is that strategies based on spectral signal analysis (CIS, SMSP, SPEAK/ACE) outperformed explicit speech-feature extraction Despite differing strategies and electrodes, there is no significant overall performance difference among present commercial devices Patient factors dominate outcome: for CA versus CIS, NU-6 scores correlate about r=0.92, so roughly 85% of outcome variance is the patient, not the strategy.[2008][2006]

TTwo organising axes

Axis 1 (waveform type): continuous analog waveforms (CA, SAS) versus interleaved biphasic pulse trains (CIS, n-of-m, ACE, SPEAK, HiRes) Axis 2 (channel selection): stimulate all channels every frame (CIS) versus peak-pick only the n largest of m channels (n-of-m, ACE, SPEAK) Nonsimultaneous (interleaved) stimulation removes the dominant channel-interaction mechanism: vector summation of overlapping electric fields at sites of neural excitation When n equals m, SPEAK/ACE reduce essentially to CIS.[2008][1991]

FWhat this chapter adds beyond the basics

This chapter recovers the feature-extraction / formant-tracking era (F0/F2, F0/F1/F2, MPEAK) that modern textbooks largely skip Every strategy is presented as a signal-flow block diagram from microphone to electrode The narrative spine is the explicit 'why each strategy superseded the last' story rather than a static catalogue The number of analysis filters and the number of stimulation channels are distinct design choices, not the same number.[1999][2010]

TBy the numbers

Across the whole lineage of strategies, CI users behave as though they have only ~4-8 effective spectral channels no matter how many physical electrodes fire. Quiet speech needs as few as 4; noise pushes the requirement up toward 8. Normal-hearing listeners keep gaining out to 20+ channels. This ceiling is why every strategy after CIS competes on HOW it allocates a handful of usable channels, not on adding more. Values are verified asymptote/plateau channel counts (Fishman 1997; Friesen 2001), not per-channel percent-correct.

FHear it

This is the core idea made audible. The chosen sound is split into 4 frequency bands; each band's envelope is extracted and re-imposed on a band of noise — just as a cochlear implant drives 4 electrodes. With one or two channels speech blurs; by 4–8 it becomes intelligible; more brings diminishing returns — the same plateau the charts show. Try the real sentence, or load your own clip (a WAV/MP3) to vocode actual speech. The bundled sentence is a synthesised text-to-speech sample (two standard Harvard sentences). Synthesised/processed live in your browser; raise the volume gently.

Case 14.1 · The Lineage at a Glance

An engineering trainee claims that 'a cochlear implant with 22 electrodes always delivers 22 independent channels of information.' Their mentor pushes back, pointing to decades of comparative data.

What is the best correction to the trainee's claim?

Self-assessment — Module 12 questions

Question 1

Which transition does Wilson date to the early 1990s as the central movement in CI coding history?

Question 2

For the comparison of CA versus CIS outcomes, what does the high NU-6 correlation (r≈0.92) imply?

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