How CueProof Scores Your Cues

A transparent look at CueProof's phonetic scoring — 10 weighted metrics, 7 degradation tests, and the canine hearing research that informed them.

Philosophy

Dogs don't hear phonemes — they hear acoustic features. Vowel formant patterns, syllable rhythm, onset energy, and consonant shape all reach a dog's ear as a blended acoustic signal, not as the discrete sound segments that human linguists describe.

CueProof's scoring engine models the dog's perspective, not the human's. Instead of comparing IPA transcriptions, it works with ASCII-folded acoustic proxies: consonant skeletons, vowel sequences, onset classes, and syllable counts. These features approximate what survives when a cue travels across a noisy field or echoes through an indoor hall.

This approach has limits — it cannot capture pitch contour, individual speaker timbre, or the exact spectral shape of a vowel. But for comparing cue pairs within a training vocabulary, these proxies are robust, language-agnostic, and grounded in the peer-reviewed literature on canine auditory perception.

Scoring Pipeline

Raw text Lowercase Remove diacritics ASCII fold Feature extraction Pairwise comparison Weighted blend Risk score

Similarity Metrics

Edit Distance

20%

Damerau-Levenshtein distance normalized to [0, 1]. Measures the minimum number of insertions, deletions, substitutions, and transpositions to transform one cue into another.

Vowel Similarity

15%

Compares the extracted vowel sequences of both cues. Dogs discriminate words primarily by vowel formant patterns across speakers.

Root-Gutteridge 2019

Consonant Skeleton

15%

Strips all vowels to produce a consonant-only skeleton, then compares. Captures the structural shape of a word that remains stable under noise.

N-gram Overlap

13%

Jaccard similarity of character bigrams and trigrams. Detects partial phonetic overlap even when overall structure differs.

Jaro-Winkler

10%

Transposition-aware string similarity with prefix bonus. Useful for detecting cues with similar but reordered sounds.

Suffix Match

8%

Compares the final 3 characters of each cue. Shared endings create rhyme-like similarity that dogs may confuse.

Prefix Match

7%

Compares the first 3 characters. Dogs are biased toward the onset of a word — shared prefixes are a strong confusion signal.

Andics 2016

Syllable Rhythm

5%

Compares syllable count and prosodic pattern. Dogs use rhythm as a coarse classifier for cue identity.

Onset Class

5%

Groups the first consonant by place of articulation. Cues starting with the same class share similar formant transitions.

Liberman 1954

Degraded Forms

2%

Applies 7 real-world degradation transforms and checks if any degraded pair converges. Low base weight, amplified by boosters.

Miller & Nicely 1955

Degradation Tests

Each cue pair is tested under 7 simulated real-world conditions. If any degraded form of cue A matches a degraded form of cue B, the pair receives a booster penalty.

Final consonant drop

Fast or shouted speech often clips the last consonant. "Sit" and "Sip" both become "Si-".

Vowel compression

Rapid utterance compresses adjacent vowels, reducing the acoustic distance between cues with similar vowel patterns.

Weak /h/ drop

At distance or in noise, the weak /h/ onset disappears. "Heel" and "Eel" become indistinguishable.

Unstressed syllable drop

Rapid speech drops unstressed syllables. Multi-syllable cues may collapse to their stressed core.

High-freq consonant confusion

Noise and distance mask high-frequency consonants first: s, sh, f, and th blur into each other.

Shout formant shift

Under arousal, vowels shift toward more open sounds: i shifts toward e, e toward a, o toward a.

Reverb stop masking

In reverberant indoor halls, stop consonants between vowels are masked. VStopV sequences collapse to VV.

Booster Penalties

After the weighted blend, additive booster penalties are applied for specific structural patterns that increase real-world confusion risk. These stack and the final score is clamped to [0, 1.0].

+0.10

Both cues are 4 characters or fewer. Fewer acoustic features for a dog to distinguish.

+0.10

Onset primacy: dogs weight the first sounds most heavily. Identical first 3 characters is a strong confusion signal.

+0.06

Shared endings create a rhyme-like acoustic tail that may sound identical from a distance.

+0.10

Same syllable count AND identical vowel pattern. The hardest combination for a dog to distinguish.

+0.15

One cue is a subset of the other. A partial utterance of the longer cue is a valid pronunciation of the shorter one.

+0.12

Any degraded form of one cue becomes identical to a degraded form of the other.

+0.08

Both start with sibilant (s/sh) or fricative (f/th) onsets that blur together at distance.

+0.05

Both start with sonorant consonants (l/r/m/n/w/y) — below 2 kHz, outside dogs' peak sensitivity.

+0.08x

Under high-drive conditions, the degraded_max component is amplified by competition pressure.

Severity Thresholds

Final risk = weighted blend + boosters, clamped to [0, 1.0]

High risk: final score above 0.65
Moderate: final score between 0.35 and 0.65
Clean: final score below 0.35 — no collision card shown

Research References

The scoring engine's weights, degradation models, and booster thresholds are informed by the following peer-reviewed studies on canine auditory perception.

  1. 1 Root-Gutteridge, Ratcliffe, Korzeniowska & Reby (2019). Dogs perceive and spontaneously normalize formant-related speaker and vowel differences in human speech sounds. Biology Letters 15(12), 20190555. DOI
  2. 2 Higaki, Farago, Pogany, Miklosi & Fugazza (2025). Sound quality impacts dogs' ability to recognize and respond to playback words. Scientific Reports 15, 14175. DOI
  3. 3 Andics, Gabor, Gacsi, Farago, Szabo & Miklosi (2016). Neural mechanisms for lexical processing in dogs. Science 353(6303), 1030-1032. DOI
  4. 4 Heffner (1983). Hearing in large and small dogs: Absolute thresholds and size of the tympanic membrane. Behavioral Neuroscience 97(2), 310-318. DOI
  5. 5 Barber, Wilkinson, Montealegre-Z, Ratcliffe, Guo & Mills (2020). A comparison of hearing and auditory functioning between dogs and humans. Comparative Cognition & Behavior Reviews 15, 45-94. DOI
  6. 6 Liberman, Delattre, Cooper & Gerstman (1954). The role of consonant-vowel transitions in the perception of the stop and nasal consonants. Psychological Monographs 68(8), 1-13. DOI
  7. 7 Mallikarjun, Shroads & Newman (2019). The cocktail party effect in the domestic dog (Canis familiaris). Animal Cognition 22(3), 423-432. DOI
  8. 8 Nabelek & Pickett (1974). Monaural and binaural speech perception through hearing aids under noise and reverberation. Journal of Speech and Hearing Research 17(4), 724-739. DOI
  9. 9 Miller & Nicely (1955). An analysis of perceptual confusions among some English consonants. Journal of the Acoustical Society of America 27(2), 338-352. DOI
  10. 10 Summers, Pisoni, Bernacki, Pedlow & Stokes (1988). Effects of noise on speech production: Acoustic and perceptual analyses. Journal of the Acoustical Society of America 84(3), 917-928. DOI

Ready to check your cues?

Try the free 3-cue collision check