ITR Atlas
← All interfaces

Accurate Speech BCI (Card et al., 2024)

Measured by Card, Stavisky, Brandman et al. · N. Engl. J. Med. 391 (2024)

Intracortical Speech invasive 2024

Inputs

The measured or assumed values behind the calculations, each with its source.

N = 125000
125,000-word vocabulary (Abstract)
rate = 32 word/min
Self-paced conversational rate sustained over 248+ hours (Abstract). Faster home-use sessions reached higher rates.
P = 0.9734
1 − 2.66% average word error rate on the 125,000-word vocabulary (as low as 1% on the best days; Abstract)

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.

Strictest Shannon (text) Recomputed
Word-entropy throughput
Effective bits actually transmitted as English
2.6 bits/s

  1. Error-corrected words per minute

    (1 − WER) × rate = 0.9734 × 32 = 31.1 net word/min

  2. Shannon per-word entropy of English

    H ≈ 5.0 bits/word

    Credits only the information in the English produced, independent of vocabulary size.

  3. Information transfer rate

    31.1 word/min × 5.0 bits/word ÷ 60 s/min = 2.6 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
Context-dependent (the live set changes per step)
Size
125000 distinguishable actions
Prior
Context-conditioned: likelihoods depend on prior actions
Notes
Same large-vocabulary, language-model-mediated structure as other free-speech BCIs: 125k words nominally, but context prunes and reweights candidates each step. The standout result is accuracy (≈97.5% sustained), not raw rate. The uniform-prior Wolpaw calculation is kept as a secondary comparison because it is not directly comparable to the atlas English-output convention.

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.

Wolpaw Recomputed
Wolpaw mutual information over N = 125,000 words
Per-word mutual information under uniform-prior Wolpaw assumptions
8.7 bits/s

  1. Bits per selection (Wolpaw formula)

    B = log2(N) + P*log2(P) + (1-P)*log2((1-P)/(N-1))
  = log2(125000) + 0.9734*log2(0.9734) + 0.0266*log2(0.0266/124999)
  = 16.304 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.

  2. Selections per second

    T = 1.875 s/selection  ->  1 / 1.875 = 0.533 selections/s

  3. Information transfer rate

    ITR = B * selections/s = 16.304 * 0.533 = 8.696 bits/s
Nuyujukian Recomputed
Nuyujukian achieved bitrate over the 125,000-word vocabulary
Large-vocabulary capacity view, shown for comparison
8.55 bits/s

  1. Achieved-bitrate credit per net-correct word

    N = 125,000 → log2(N − 1) = log2(124,999) ≈ 16.93 bits per net-correct selection (Nuyujukian 2015, which introduced the metric, used log2(N); at this N the difference is negligible).

  2. Net-correct word rate

    net-correct = 2P − 1 = 2(0.9734) − 1 = 0.947 of words. At 32 word/min (1.875 s/word) → 0.947 × 32 / 60 = 0.505 correct/s.

    A word error commits the wrong word rather than timing out, so incorrect = 1 − P. Same N (125,000), word accuracy (97.34%) and rate (32 wpm) as the entry's Wolpaw calc. As that calc's own caveat notes, the 125k-word vocabulary is context-reweighted by the language model each step, so feeding it into log2(N − 1) is a large-vocabulary capacity view (~17 bits/word) — raw channel capacity, not communication, analogous to the nagel code-space figure. The ranked communication rate is the 2.6 bits/s word-entropy Shannon.

  3. Achieved bitrate

    16.93 bits × 0.505 correct/s = 8.55 bits/s.

Source

Authors
Card, Stavisky, Brandman et al.
Publication
N. Engl. J. Med. 391, 2024
Paper
10.1056/NEJMoa2314132
Reference
Open-access full text (PMC)