KKmatch/KKOS AI/Case results
Case results

Measured on real speech, not asserted.

Everything below comes out of one offline script that anyone can re-run. Where a layer has not been measured yet, it says so.

01 · Acoustic tracing

Three real recordings, traced frame by frame

The engine reads 40 ms frames with a 20 ms hop. Below are the aggregates it returns for each recording — no smoothing, no rounding up.

Pipeline frame-autocorr-unbiased/v2 · provider numpy-autocorr · jitter and shimmer are frame-level proxies and must not be compared numerically against Praat period-based figures. The two aggregates that move under padding are the ones that divide by total duration.

02 · Reconstruction

How few numbers does a spoken line need?

We take the measured pitch track, then rebuild it from K straight segments and measure the residual. The optimum is found by dynamic programming, so the error can only fall as K rises.

Pitch-contour reconstruction error

Two recordings, four budgets each. Vertical axis is error in semitones; horizontal axis is how many segments the contour was compressed to.

0.00 sterror at K = 32 · short sentence
0.00 sterror at K = 16 · short sentence
Recording K = 4 K = 8 K = 16 K = 32
TTS short sentence11.586.673.381.08
End-to-end session capture41.7738.7332.7722.25

Unit is the semitone. The session capture needs far more budget because it contains many more distinct pitch targets — that is a property of the recording, not a defect of the estimator. Frame-level pitch tracks are used inside the measurement script only and are never stored or published.

03 · Separation

Five intended expressions, five recovered

We built five carriers whose pitch level and pitch variability were fixed by design, then let the engine measure them blind. The decision rule was written down before the run.

5 / 5 intended expression recovered hand-written rule, no training thresholds pre-registered ring = 1 − pitch-spread error
What this does not show Five synthetic carriers are not five people. This measures whether the engine can separate prosodic patterns that were deliberately made separable — it is not a human emotion-recognition accuracy, and we do not publish it as one. A real accuracy figure needs labelled human audio, which does not exist in this build.
04 · Inversion

Reading the design back out of the measurement

For each carrier we know the pitch level and pitch spread we asked for. Here is what the engine reported back, and the gap.

0.000%mean pitch-level error
0.00%mean pitch-spread error
Intended expression Designed f0 Measured f0 Error Designed spread Measured spread Error
Understated162.0161.93−0.04%0.0900.092+2.3%
Calm, unhurried135.0135.07+0.05%0.1000.107+6.5%
Low and heavy112.0110.96−0.93%0.1200.155+29.2%
Tense, quick205.0203.03−0.96%0.2600.279+7.1%
Flat160.0160.22+0.14%0.0350.036+1.4%

The worst case is the lowest carrier, and the reason is arithmetic: at 112 Hz a 40 ms frame holds only about 4.5 periods, so the autocorrelation peak is positioned with more variance than at 205 Hz, where the same frame holds about 8. The mean level is recovered far better than the spread, and we report the spread error rather than hide it behind an average.

05 · Status

What is measured, and what is not

A results page is only worth reading if it lists its own gaps. These are ours, as of this build.

LayerStateBasis
Pitch estimation on synthetic tones39 / 39140 Hz–300 Hz sweep, 13 assertions, all passing
Acoustic tracing on real speechmeasured3 recordings, aggregates only
Pitch-contour reconstructionmeasuredDP-optimal piecewise-linear fit
Expression separationmeasured5 synthetic carriers, pre-registered rule
Human-labelled audio agreement0no consented audio, no annotators yet
Compressed-audio decodingoff locallyPyAV not installed on the measurement host
Not measured Human agreement is the one that matters most and it is empty. Voice is a sensitive biometric: any audio used to calibrate the engine needs an explicit consent record and is used once, never stored or redistributed. Until that pipeline has real consented recordings behind it, the honest number is zero.
06 · Reproduce

One script, one seed

Every figure on this page is written to a single JSON file by a single offline script. Nothing on this page carries its own data.