SilentSpeller (Kimura et al., 2022)
Measured by Kimura, Gemicioglu, Womack, Zhao, Li, Bedri, Su, Olwal, Rekimoto & Starner · ACM CHI 2022 (2022)
Inputs
The measured or assumed values behind the calculations, each with its source.
- N = 1164
- 1,164-word dictionary the speller decodes against
- P = 0.87
- Live word accuracy averaged over 7 participants, at 37 wpm. Offline isolated-word character accuracy was 97% on the dictionary; an earlier eval reached 53 wpm at 90%.
- rate = 37 word/min
- Live text-entry speed, 7-participant average (a real measured rate, unlike most silent-speech recognizers)
Strictest ITR
Each scoring method is an upper bound on the channel, so the headline is the strictest (smallest) one for this entry. Use the score selector on the home page to view any single method across entries.
-
Error-corrected words per minute
(1 − WER) × rate = 0.87 × 37 = 32.2 net word/min
37 wpm is a real measured live rate, unlike the assumed-rate silent-speech entries.
-
Shannon per-word entropy of English
H ≈ 5.0 bits/word
-
Information transfer rate
32.2 word/min × 5.0 bits/word ÷ 60 s/min = 2.68 bits/s
What counts as a bit depends on the action space. The number of distinguishable actions and how likely each one is are design choices of the task, not the sensing hardware. The same modality can present a fixed set of targets, a set pruned per step by a grammar or language model, or a continuous control space. Each of these changes how many actions are live and how the probability mass is spread, and therefore the information per selection. Read the action space below before comparing headline numbers across entries.
Action space
What the user can produce at each step, and how those options are distributed.
- Structure
- Fixed set of targets
- Size
- 1164 distinguishable actions
- Prior
- Context-conditioned: likelihoods depend on prior actions
- Notes
- An oral wearable (dental retainer with capacitive touch sensors) tracks tongue contact to spell words letter by letter, decoded against a 1,164-word dictionary. The dictionary constraint plus a language model make the per-word prior context-conditioned rather than uniform.
Comparability The strictest bound here is the Shannon entropy of the output text, under one predictor held constant across the whole atlas (≈1 bit per character). That shared predictor makes it directly comparable to every other text entry (keyboards, spellers, silent speech and speech BCIs) regardless of prior or vocabulary size. For most text interfaces it comes out tighter than the raw-selection bounds, but not always. Where a small vocabulary makes Wolpaw tighter, that wins instead. Any Fitts, Wolpaw or log₂(N) figure shown below is another bound on the same channel. Switch the home-page score selector to compare one across entries.
Other score types
Bounds the atlas keeps out of the default strictest headline: as-reported figures, alternate task conditions, or raw-channel ceilings that shouldn't win the headline by default. Each still carries a score type, so the home-page selector ranks this entry on it when you choose that type. Read its derivation before comparing across entries.
-
Bits per selection (Wolpaw formula)
B = log2(N) + P*log2(P) + (1-P)*log2((1-P)/(N-1)) = log2(1164) + 0.87*log2(0.87) + 0.13*log2(0.13/1163) = 8.304 bits / selection
Term 1 is the information if every choice were correct; terms 2-3 subtract the bits lost to the error rate, assumed spread evenly over the other N-1 targets.
-
Selections per second
T = 1.62162 s/selection -> 1 / 1.62162 = 0.617 selections/s
-
Information transfer rate
ITR = B * selections/s = 8.304 * 0.617 = 5.121 bits/s
-
Achieved-bitrate credit per net-correct word
N = 1,164 → log2(N − 1) = log2(1163) = 10.18 bits per net-correct selection (field-standard achieved bitrate, e.g. Webgrid; Nuyujukian 2015, which introduced the metric, used log2(N)).
-
Net-correct word rate
net-correct = 2P − 1 = 2(0.87) − 1 = 0.74 of words. At the measured 37 word/min (1.62 s/word) → 0.74 × 37 / 60 = 0.456 correct/s.
A word error commits the wrong word rather than timing out, so incorrect = 1 − P. Same N (1,164), live word accuracy (87%) and measured rate (37 wpm) as the entry's Wolpaw calc. The dictionary is language-model-constrained, so this is a per-word capacity view over the 1,164-word set, not open-vocabulary throughput; the ranked figure is the 2.68 bits/s Shannon. Rate is measured, unlike the assumed-rate silent-speech entries.
-
Achieved bitrate
10.18 bits × 0.456 correct/s = 4.65 bits/s.
Source
- Authors
- Kimura, Gemicioglu, Womack, Zhao, Li, Bedri, Su, Olwal, Rekimoto & Starner
- Publication
- ACM CHI 2022, 2022
- Paper
- 10.1145/3491102.3502015
- Reference
- Author copy (PDF)