61 tiny MLP regression heads on top of the frozenlaion/voiceclap-commercial
768-d speech embedding. One head per attribute; each head reads a single 768-d
embedding and emits that attribute's continuous score directly — no calibration
step, no bucket decoding.
The heads cover 55 perceptual/emotional voice dimensions (Anger, Valence,
Arousal, Warm_vs._Cold, …), 4 audio-quality scores, plus Age, Gender,
duration and talking_speed.
External benchmarks are reported below — these are validation
numbers on the training distribution, and they do not transfer unchanged.
Read this before you use the numbers
These four caveats are not footnotes; they change how the table should be read.
1. duration r = 0.998 is trivial and is not evidence of embedding quality
Waveform length is directly visible to the encoder, so predicting it is close to
reading it off. It also saturates at 30 s, because that is the encoder window
used everywhere in this project (the training target was clipped to 30 s as well).
It is included as a pipeline sanity check, nothing more. Ignore it when judging how
good the representation is.
2. Bucket-balanced training pushes the predicted mean up on rare emotions
Training sampled inversely to bucket frequency so that rare high-intensity examples
were not drowned out. On zero-inflated dimensions this systematically shifts the
predicted level upward. The worst offenders (see the bias = mean(pred) − mean(true)
column): Thankfulness_Gratitude+0.65, Relief+0.49, Distress +0.44,
Impatience_and_Irritability +0.42, Awe +0.41, Astonishment_Surprise +0.40,
Infatuation +0.40, Disappointment +0.39, Anger +0.39, Triumph +0.38,
Affection +0.38, Sadness +0.37, Fear+0.35.
Ranking within a dimension is good; the absolute level runs high. If you need a
calibrated absolute value, subtract the bias column, or (better) fit a one-off
affine map on a small sample of your domain. Do not threshold a raw score at "0.5
means present" without recalibrating.
3. On sparse emotions, Spearman ρ is the stricter and more honest read
Pearson r on a zero-inflated target is inflated by the easy separation of "lots of
zeros" from "a few large values". Where r and ρ disagree, trust ρ:
Anger r = 0.728 but ρ = 0.516; Malevolence_Malice 0.679 → 0.466;
Distress 0.481 → 0.289; Pain 0.392 → 0.199; Sadness 0.503 → 0.345.
4. talking_speed is characters per second, not words per second
It reproduces the source dataset's talking_speed_5.00_to_25.00 bucket definition.
Typical English conversational speech lands around 12–17 chars/s. Divide by ~5.5 for
a rough words-per-second figure.
Reproducibility note that matters more than any hyperparameter: de-duplicate before you split
The training dataset files each clip under one WebDataset tar per dimension, so a
single clip appears in up to 15 different tars. The raw tar-entry count is
504,007; the number of unique clips is 296,422.
If you build the train/val split on raw tar entries, the same clip lands in both
train and validation, and every metric in every table becomes inflated — the model
is being asked to recall an embedding it has already memorised. De-duplication is done
here by WebDataset key, before the split, and it is the single most important step
to reproduce these numbers rather than better-looking fake ones.
(Related quirk handled the same way: the Jealousy___Envy_* and Jealousy_and_Envy_*
tars both carry the single score key Jealousy_&_Envy, so those two near-duplicate tar
dimensions merge into one head. That is why 328 tars over ~15 dimension families
yield 61 — not 62 — heads.)
Usage
Both files below are in this repo: voiceclap_heads.py is a
~180-line reference implementation, example_inference.py is
the demo whose real output is pasted underneath.
1paths =["a.wav","b.wav","c.wav"]23# all 61 heads, or a subset:4out = scorer.score(paths, dims=["Valence","Arousal","Anger"], batch_size=32)5# {'Valence': array([...]), 'Arousal': array([...]), 'Anger': array([...])}67# or keep the embeddings and re-run heads later (they are the expensive part):8import numpy as np
9from voiceclap_heads import load_audio
10emb = scorer.embed([load_audio(p)for p in paths], batch_size=32)# (N, 768)11scores = scorer.score_embeddings(emb)# 61 x (N,)
Inference contract (must match, or the numbers do not transfer)
mono, resampled to 16 kHz
truncated to 30 s (480,000 samples) — the encoder window
batch padded to that batch's max length, not a fixed 30 s
(voiceclap_heads.AttributeScorer.embed sorts by length and does this for you;
padding policy has a small but real effect on the embedding, so keep it identical)
VoiceClap.encode_waveform(...) → (N, 768)
x = (emb - emb_mu) / emb_sd — both vectors ship inside heads.pt
head(x) → the score. The y-normalisation affine is already folded into each
head's last Linear, so there is nothing to de-standardise afterwards.
Score scales and polarity
Scores are on the source annotation scale, which is roughly 0–7 for emotions
(0 = not present) — but each dimension has its own empirical range. report/results.csv
carries label_min, label_max, label_mean, label_p50, label_p99 and
frac_below_0p5 per dimension; config.json carries y_min/y_max/y_mean/y_std
per head. Heads are unbounded regressors — clip to the label range if you need it.
Polarity of the non-obvious dimensions, determined empirically (not assumed) by
correlating our heads against the independent
laion/voicenet-dimension-predictors-commercial
predictors on 61 held-out clips:
dimension
direction
evidence
Gender
higher = more feminine
r = −0.90 vs VoiceNet GEND (whose level 0 = "hyper-feminine", level 6 = "hyper-masculine")
Age
higher = older
r = +0.52 vs VoiceNet AGEV (level 0 = infant → 6 = elderly)
X_vs._Y names
higher = more Y
e.g. High-Pitched_vs._Low-Pitched high ⇒ lower-pitched; Monotone_vs._Expressive high ⇒ more expressive
duration
seconds, capped at 30
talking_speed
characters/second
Background_Noise
higher = more noise
r = −0.29 with score_background_quality, −0.46 with Recording_Quality; ρ = +0.60 against emolia's BKGN rubric (whose level 6 = "noise overpowers the voice")
Run on this machine with python example_inference.py examples/clip_a.mp3 examples/clip_b.mp3
(CPU, ~2.7 s for both clips after model load). The two clips ship in
examples/. They are LAION-generated studio-style speech and were not
in the training set.
clip_a.mp3 — generated from the direction "A 35-year-old female with a warm,
resonant, slightly husky timbre … relaxed, a subtle smile playing on her lips", 22.15 s
clip_b.mp3 — generated from the direction "A 78-year-old man with a gravelly,
slightly strained voice … weary but resolute", 24.07 s
Reading it honestly: duration is recovered to within 0.12 s / 0.17 s; Age and
Gender order the two speakers the right way (older & more masculine for clip_b);
the older, weary clip gets much lower Arousal (1.65 vs 3.21) and lower Valence.
Age 2.587 vs 3.199 is a 7-point scale, not years — it is not claiming 35 vs 78.
A caution on that example. The prompt text is a generation direction, not a
measurement of the produced audio. On a 61-clip sample of the same pool, our Gender
head correlates r = 0.09 with the gender stated in the prompt and Age correlates
r = 0.05 with the stated age — while agreeing strongly with VoiceNet's independent
predictors on the same audio (|r| = 0.90 and 0.52). The synthesis simply does not
reliably follow the prompt's speaker description on this pool, so prompts are not usable
as ground truth. Use the examples to check that your pipeline is wired correctly, not as
an accuracy claim.
Targets: every sample's JSON carries the full 59-score
empathic_insight_scores / annotation_scores vector, so every clip labels every
dimension and the regression targets are exact continuous scores — not bucket
midpoints reconstructed from the tar name. That is why n_train is ~266.8 k for
essentially every head. Plus duration and talking_speed derived from the JSON.
Embeddings: computed once with the frozen encoder into a single
(296422, 768) float16 array; the heads then train in minutes.
Loss: Huber (δ tuned per dimension over {0.5, 1.0}).
Split: 10 % validation, stratified per score bucket, taken after de-duplication.
Sampling: bucket-balanced (weight ∝ 1/bucket count) — see caveat 2.
Feature standardisation: per-dimension mean/std over a 200 k random subsample,
shared by all heads (emb_mu, emb_sd in heads.pt).
Sweep: 19 runs per dimension — {mlp1_h64, mlp1_h128, mlp2_h96_64, mlp2_h64_32}
× lr {1e-3, 3e-3} × wd {1e-4, 1e-2} (16 runs), then 3 dropout / Huber-δ refinement
runs on the winning architecture. Selected by validation Pearson r, with cosine
schedule, 200-step warmup, early stopping (patience 8 evals).
EmoNet Voice Bench evaluation: script, per-row predictions, full metrics JSON, label table, data manifest
provenance/meta_all.parquet
one row per unique clip: key, tar, dim, bucket, src, dur, nchar, nword — enough to reproduce the de-duplication and the split
The 296,422 × 768 embedding matrix and the 296,422 × 61 label matrix are not
uploaded (0.5 GB). Regenerate them from the dataset with the contract above, or ask.
Intended use and limits
Fast, cheap attribute tagging, dataset filtering/curation, TTS-output QA, retrieval and
reward shaping over large speech corpora — anywhere you want 61 numbers per clip at
roughly the price of one embedding.
Limits.
Training labels are model-generated (Empathic Insight annotations), not human
ratings. These heads distil an annotator model; they inherit its biases, and agreement
with human raters is an open question — see the benchmark section.
Training audio is largely synthetic / TTS-generated English studio-style speech.
Expect degradation on real-world noisy recordings, on non-English speech, and on
spontaneous conversational audio.
Medium-band emotions (r 0.3–0.6) are usable for ranking and coarse filtering, not
for per-clip assertions about a speaker's emotional state.
Do not use these to make consequential judgements about individuals (hiring,
policing, clinical, insurance). Gender and Age are perceived-voice attributes, not
identity claims, and are wrong often enough to be harmful if treated otherwise.
Citation
bibtex
1@misc{voiceclap_attribute_heads_2026,
2 title = {VoiceClap Attribute-Regression Heads: 61 linear-probe dimensions over a frozen commercial speech encoder},
3 author = {LAION},
4 year = {2026},
5 url = {https://huggingface.co/laion/voiceclap-commercial-attribute-heads}
6}
Base encoder: laion/voiceclap-commercial. Training data:
TTS-AGI/Emotion-Voice-Attribute-Reference-Snippets-DACVAE-Wave.
Benchmarks
Two external benchmarks were run against these heads. Neither is a regression
benchmark shaped exactly like our training task, so read the methodology before
the numbers.
Task, verbatim from the dataset. Each row carries exactly one target emotion
and an intensity rated 0 = Not Present / 1 = Mildly Present / 2 = Intensely Present
by 2–4 of 6 psychology experts. It is not a 40-way argmax and not multi-label.
The paper maps the averaged human intensity to a 0–10 scale (0 → 0, 1 → 5, 2 → 10) and
reports MAE, RMSE, Pearson r and Spearman ρ.
Our protocol. For each row we take the single head named by that row's emotion
and score the clip with it — no argmax, no head selection. 42 of 42 benchmark classes
map 1:1 onto a head, with only cosmetic renames
(Astonishment→Astonishment_Surprise, Fatigue→Fatigue_Exhaustion,
Hope→Hope_Enthusiasm_Optimism, Intoxication→Intoxication_Altered_States_of_Consciousness,
Malevolence→Malevolence_Malice, Pleasure→Pleasure_Ecstasy,
Thankfulness→Thankfulness_Gratitude, Emotional Numbness, Jealousy & Envy,
Sexual Lust, Impatience and Irritability). The remaining 19 heads are unused.
All 12,600 rows were scored; refusal rate 0 %.
Results
aggregation
Pearson r
Spearman ρ
pooled over all 12,600 rows
0.304
0.323
macro-averaged over the 42 emotions
0.386
0.374
macro, restricted to the 2,912 rows with unanimous annotators
0.548
0.488
Published baselines (paper Table 5), for orientation:
model
Spearman ρ
Pearson r
MAE
RMSE
refusal
EmpathicInsight-Voice Large
0.415
0.421
2.995
3.756
0 %
EmpathicInsight-Voice Small
0.418
0.414
2.997
3.757
0 %
Gemini 2.5 Pro
0.417
0.416
3.008
3.785
0 %
Gemini 2.0 Flash
0.355
0.350
3.608
4.453
0.01 %
GPT-4o Audio Preview
0.337
0.336
3.432
4.247
27.6 %
GPT-4o Mini Audio Preview
0.326
0.327
3.320
4.124
2.3 %
Hume Voice
0.274
0.231
4.744
5.474
39.2 %
these heads (pooled)
0.323
0.304
see below
see below
0 %
Honest reading: on the fair, scale-free comparison these 61-head probes land
below Gemini 2.0 Flash and the EmpathicInsight-Voice models, above Hume Voice, and
roughly level with GPT-4o Mini Audio Preview — while being a 50 k-parameter MLP on a
frozen embedding rather than a large audio LLM. The paper does not state whether its
Table 5 aggregates pooled or macro-over-emotions; we give both, and we quote our
pooled figure in the table above because it is the less flattering of the two.
MAE / RMSE need calibration and are therefore not comparable
Our heads emit the Empathic-Insight 0–7 scale, not 0–10, and the benchmark's rows are
heavily emotion-present (gold mean 5.26 on the 0–10 scale) while our raw predictions
average 1.13. Raw MAE is dominated by that scale mismatch, not by ranking quality:
variant
MAE
RMSE
fair to compare?
fixed rescale clip(score/7×10, 0, 10), no fitting
3.884
4.587
yes, but it is mostly measuring the scale gap
one global affine, 2-fold cross-fitted on the benchmark
2.289
2.786
no — uses benchmark labels
per-emotion affine, 2-fold cross-fitted on the benchmark
2.142
2.607
no — uses benchmark labels
per-emotion affine fit on all data (oracle)
2.124
2.579
no — upper bound only
The calibrated rows would "beat" every model in Table 5. They are not a win. The
LLM baselines are zero-shot and receive no in-domain calibration; we do. We report them
only to show that most of the raw MAE gap is an offset/scale artefact rather than a
ranking failure. Use the correlation table for comparisons. This is caveat 2 (the
bucket-balanced bias) showing up in a second domain, in the opposite direction: our
absolute level is far too low here because the benchmark over-samples present-emotion
clips relative to our training marginal.
Presence detection (a secondary, non-official metric)
Reframing as "is this emotion present at all?" — clips all annotators rated 0 vs clips
all annotators rated > 0 — gives a macro ROC-AUC of 0.810 over the 38 emotions with
≥ 10 clips on both sides. This is not the benchmark's defined metric; it is reported
because it is the shape most downstream users actually want (filtering a corpus), and
because it is threshold-free.
Contamination check
Identifier spaces are disjoint: benchmark clips are named <8-hex>_enhanced*.mp3,
our training keys are cond_podcastt_* / emo_<dim>_b<bucket>_batch*_chunk*. Zero
overlap on exact stems and zero overlap on 8-hex tokens. That is evidence of separate
provenance, not proof of zero audio overlap — no audio-fingerprint dedup was run.
A lineage caveat that is more important than the filename check: our training
labels are Empathic Insight model annotations, and the EmpathicInsight-Voice models
were themselves trained on EmoNet-Voice data. So our teacher may have seen clips from
the same parent corpus as this benchmark. Treat these numbers as distillation
transfer measured against human experts, not as a clean held-out generalisation test
against an unrelated corpus. The gap between our 0.30 and EmpathicInsight-Voice's 0.42
is roughly the cost of distilling that annotator into a 50 k-parameter probe.
Per-emotion results
presence AUROC is blank where one side had < 10 clips.
Benchmark:LAION-AI/emolia-bench
(commit 1a48a7a). Two subsets, both binary classification of (audio, text-prompt)
pairs against majority_present from 2–4 human raters:
emolia-dim — 18,632 items over the 58-dimension VoiceNet rubric
(58 dims × 7 levels × {positive, negative}).
Official metric: balanced accuracy vs majority_present
(benchmark.py::binary_metrics, bal = (sensitivity + specificity)/2); secondary
metric positive-class F1. We ran the benchmark's own harness through its
--endpoint path — a local server returns the relevant head's score as the
"similarity" — with the README's recommended filter --min-raters 2 --exclude-flagged.
2a. emolia-emo — ran in full, 40/40 emotions map to a head
n = 7,986 after filtering (2 items dropped for having < 2 raters;
--exclude-flagged dropped 0, because this subset's benchmark_labels.csv has no
flagged column at all — that filter is a silent no-op here, not evidence of clean
audio). Positive rate 0.578. 0 missing audio, 0 evaluation errors.
Only one name needed mangling: the benchmark's Jealousy / Envy → head Jealousy_&_Envy.
21 of the 61 heads are unused.
Threshold. The harness defaults to --threshold 0.0, which assumes a cosine-style
score centred on zero. Our heads emit a non-negative 0–7 intensity, so that default is
degenerate (balanced accuracy 0.515 — it calls almost everything present). We used
--threshold 1.0, chosen a priori as the Empathic-Insight "mildly present"
boundary, with no reference to benchmark labels. For transparency: the oracle-tuned
threshold is 0.919 and yields 0.696, so the a-priori choice cost ~0.002.
balanced accuracy
F1
all 7,986 items, threshold 1.0 (headline)
0.694
0.696
unanimous-rater subset (n = 2,614)
0.822
0.867
harness default threshold 0.0
0.515
0.734
oracle-tuned threshold 0.919 (not a headline)
0.696
0.713
Baselines the benchmark itself computes on the identical task:
reference
balanced accuracy
random / always-majority
0.500
single human, pairwise
0.562
single human, leave-one-out vs the majority of the others (n = 13,208)
0.572
these heads
0.694
By the benchmark's own rubric that is the "Medium — above single-human, useful
training target" band.
The 0.694 is inflated, and here is the measurement that shows it
emolia-emo's item pool was selected using an earlier version of these very heads:
every sidecar JSON under dataset/emolia-emo/data/<Emotion>_best/ carries 55
<Dimension>_best columns from this head family, and the folder name means "top-scoring
clips for that head". The human labels are independent, so this is selection bias,
not label leakage — but it is home turf, and the pooled number is not a clean estimate.
The benchmark's 5 task_type strata let us measure how much. Within a stratum the
selection confound is held roughly constant:
task_type
n
balanced acc
AUROC
affirmative
3,999
0.629
0.706
contrastive_1
1,000
0.535
0.559
contrastive_2
1,000
0.539
0.556
penultimate
992
0.526
0.562
ultimate
995
0.501
0.576
pooled
7,986
0.694
0.748
mean within stratum
0.546
0.592
Roughly half the pooled signal is the selection confound. The honest headline is the
mean within-stratum AUROC of 0.592 — modest, above chance, and far below what the
pooled 0.748 suggests. On the hard contrastive/ultimate strata — where the negatives are
near-misses rather than random other emotions — these heads are close to chance.
A useful control: scoring the same items with the sidecar's own older head values gives
pooled AUROC 0.749 and mean within-stratum AUROC 0.573, versus 0.748 / 0.592 for the
heads published here. The released heads are marginally better and behave essentially
like the selector, which is exactly what the selection-bias story predicts.
Note also that the benchmark's own inter-rater agreement is low (Fleiss κ ≈ 0.086 binary
on emolia-emo; single-human LOO balanced accuracy 0.572), so there is little headroom
above ~0.6 for anyone on the within-stratum reading.
2b. emolia-dim — not run as a headline number, and here is why
We deliberately do not report a balanced accuracy for emolia-dim. Three independent
reasons, any one of which would be disqualifying:
Half the benchmark has no corresponding head. 30 of 58 dimensions — 9,360 rows,
50.2 % — have no head at all: ARSH, ATCK, CHNK, DARC, DFLU, EMPH, EXPL, FULL, LANG, METL, RESP, R_CHST, R_HEAD, R_MASK, R_MIXD, R_NASL, R_ORAL, R_THRT, SMTH, STRU, S_CONV, S_MONO, S_NARR, S_NEWS, S_STRY, S_TECH, VALS, VFLX, VOLT, WARM.
Some are structurally unreachable, not merely missing. The 7 resonance dimensions
(R_CHST/THRT/ORAL/MASK/NASL/HEAD/MIXD) and the 6 within-clip dynamics dimensions
(VOLT, VFLX, DARC, ARSH, VALS, ATCK) describe trajectory inside the clip. Our
heads emit one clip-level scalar. No calibration fixes that; it needs a windowed head.
The task shape does not fit a monotone scalar.emolia-dim asks "does this clip
sit at level L?", not "how much of D?". A monotone score thresholded once
globally cannot answer bucket membership. (This is the same reason CLAP baselines
struggle on this subset — the harness was built for cosine similarity.)
WARM deserves a specific flag: it is a false friend. The rubric defines it as
spectral warmth (200–500 Hz low-mid energy); our Warm_vs._Cold is affective /
interpersonal warmth. Same word, different construct — so we list it as "no head" rather
than claim the match.
What we do report — a diagnostic, not the benchmark's metric. For the 8 dimensions
that genuinely share a construct with a head, Spearman ρ between the head score and the
queried rubric level, restricted to items humans agreed sit at that level. This measures
monotone agreement with the rubric's ordering:
VoiceNet dim
head
sign
n
Spearman ρ
GEND Perceived Gender
Gender
−
66
0.681
BKGN Background Noise
Background_Noise
+
79
0.597
VALN Valence
Valence
+
89
0.589
STNC Stance
Submissive_vs._Dominant
+
97
0.518
AROU Arousal
Arousal
+
75
0.450
TEMP Tempo
talking_speed
+
64
0.441
RCQL Recording Quality
Recording_Quality
+
78
0.430
VULN Vulnerability
Vulnerable_vs._Emotionally_Detached
−
89
0.302
macro mean
0.501
Sign conventions were read off the rubric text, not fitted: GEND level 6 =
hyper-masculine while our Gender rises toward feminine; VULN level 0 = "impenetrable,
heavily armored" and level 6 = "profoundly raw, naked", while our head rises toward
Emotionally Detached. One trap worth recording: emolia's BKGN rubric runs opposite
to the one in voicenet .../dimensions.py — here level 0 is a "dead acoustic vacuum"
and level 6 is "noise overpowers the voice", so +1 is correct for this rubric and
would be wrong for the other. Sample sizes are small (64–97 items per dimension); treat
these as indicative.
As an additional sanity figure only: applying a label-free quantile-matching
calibration (map each head score onto 0–6 by matching its empirical quantiles to the
level marginal, then predict "yes" iff the mapped level equals the queried level) gives a
macro balanced accuracy of 0.548 over those 8 dimensions. It covers 8 of 58
dimensions and is not comparable to a full-benchmark number.
Artifacts, scripts and the harness's own report.md / summary.json are in
bench/emolia/.
Benchmarks — what to take away
On human-rated emotion-intensity regression (EmoNet Voice Bench) these probes reach
pooled r = 0.304 / ρ = 0.323 with a 0 % refusal rate: below Gemini 2.0 Flash and the
EmpathicInsight-Voice models, above Hume Voice, at ~50 k parameters per dimension.
On binary attribute presence (emolia-emo) the pooled 0.694 balanced accuracy looks
strong and beats the single-human ceiling, but roughly half of it is a selection
artefact; the defensible figure is a mean within-stratum AUROC of 0.592.
emolia-dim is not a task these heads can do, and we report no number for it.
Across both: ranking is decent, absolute calibration is not. Recalibrate on your
own domain before thresholding.
Full results (all 61 dimensions)
Held-out validation, 10 % bucket-stratified split, taken after key de-duplication.
bias = mean(prediction) − mean(target) on validation (caveat 2). Sorted by Pearson r.
#
dimension
band
arch
params
n_train
n_val
MAE
RMSE
Pearson r
Spearman ρ
bias
label range
label mean
1
duration
strong
mlp2_h96_64
80,289
266,780
29,642
0.387
0.586
0.998
0.997
+0.025
3.0 … 30.0
13.344
2
score_overall_quality
strong
mlp2_h96_64
80,289
266,779
29,643
0.088
0.114
0.949
0.941
-0.005
1.29 … 3.66
2.792
3
Gender
strong
mlp2_h96_64
80,289
266,779
29,643
0.26
0.373
0.94
0.907
+0.036
-2.666 … 2.291
-0.096
4
score_background_quality
strong
mlp2_h96_64
80,289
266,780
29,642
0.105
0.141
0.935
0.911
+0.003
1.523 … 4.523
3.463
5
score_content_enjoyment
strong
mlp2_h96_64
80,289
266,780
29,642
0.073
0.095
0.933
0.906
+0.022
3.098 … 5.883
4.877
6
Monotone_vs._Expressive
strong
mlp2_h96_64
80,289
266,780
29,642
0.214
0.288
0.901
0.888
-0.030
0.408 … 4.527
2.402
7
Serious_vs._Humorous
strong
mlp1_h128
98,561
266,779
29,643
0.266
0.359
0.896
0.876
+0.032
-0.274 … 5.016
1.507
8
High-Pitched_vs._Low-Pitched
strong
mlp2_h96_64
80,289
266,780
29,642
0.108
0.147
0.894
0.882
+0.011
0.544 … 4.461
1.782
9
Recording_Quality
strong
mlp1_h128
98,561
266,781
29,641
0.182
0.234
0.893
0.896
+0.024
0.757 … 4.656
2.471
10
Amusement
strong
mlp2_h96_64
80,289
266,780
29,642
0.338
0.441
0.89
0.859
+0.153
-0.01 … 4.672
0.865
11
Concentration
strong
mlp1_h128
98,561
266,781
29,641
0.26
0.344
0.888
0.881
+0.074
-0.003 … 4.562
0.862
12
Vulnerable_vs._Emotionally_Detached
strong
mlp2_h96_64
80,289
266,621
29,625
0.229
0.301
0.849
0.837
+0.041
-0.049 … 4.074
1.46
13
Interest
strong
mlp2_h96_64
80,289
266,779
29,643
0.25
0.351
0.834
0.817
-0.060
-0.051 … 3.65
2.024
14
Confident_vs._Hesitant
strong
mlp2_h96_64
80,289
266,781
29,641
0.251
0.343
0.824
0.789
+0.089
0.085 … 4.156
1.281
15
Arousal
strong
mlp2_h96_64
80,289
266,780
29,642
0.296
0.395
0.822
0.816
+0.024
-0.015 … 5.019
1.897
16
score_speech_quality
strong
mlp2_h96_64
80,289
266,779
29,643
0.045
0.057
0.807
0.721
+0.017
1.801 … 2.361
1.924
17
Embarrassment
strong
mlp2_h64_32
51,457
266,780
29,642
0.248
0.329
0.799
0.76
+0.138
-0.095 … 2.988
0.392
18
Submissive_vs._Dominant
strong
mlp2_h96_64
80,289
266,781
29,641
0.174
0.244
0.795
0.77
+0.047
-0.656 … 2.389
0.483
19
Age
strong
mlp2_h96_64
80,289
266,780
29,642
0.178
0.274
0.792
0.794
-0.036
0.077 … 5.418
2.961
20
Teasing
strong
mlp2_h64_32
51,457
266,780
29,642
0.308
0.379
0.777
0.753
+0.137
-0.012 … 3.662
0.548
21
Soft_vs._Harsh
strong
mlp2_h96_64
80,289
266,779
29,643
0.223
0.323
0.754
0.721
-0.062
-1.956 … 1.758
0.439
22
Background_Noise
strong
mlp1_h128
98,561
266,782
29,640
0.18
0.234
0.753
0.746
-0.010
-0.036 … 2.131
0.549
23
Contemplation
strong
mlp2_h96_64
80,289
266,780
29,642
0.357
0.483
0.748
0.726
+0.241
-0.039 … 4.184
0.504
24
Anger
strong
mlp2_h96_64
80,289
266,780
29,642
0.491
0.6
0.728
0.516
+0.388
-0.02 … 6.141
0.371
25
talking_speed
strong
mlp1_h128
98,561
266,781
29,641
2.592
3.683
0.726
0.75
-0.472
0.124 … 100.694
14.908
26
Warm_vs._Cold
strong
mlp1_h128
98,561
266,562
29,617
0.314
0.438
0.724
0.699
-0.001
-2.215 … 2.738
0.476
27
Authenticity
strong
mlp1_h128
98,561
266,780
29,642
0.132
0.169
0.718
0.701
+0.026
1.594 … 4.176
2.918
28
Impatience_and_Irritability
strong
mlp2_h96_64
80,289
266,780
29,642
0.563
0.691
0.718
0.598
+0.419
-0.075 … 5.707
0.533
29
Intoxication_Altered_States_of_Consciousness
strong
mlp2_h96_64
80,289
266,780
29,642
0.383
0.53
0.696
0.61
+0.279
-0.111 … 4.562
0.249
30
Pleasure_Ecstasy
strong
mlp2_h64_32
51,457
266,780
29,642
0.333
0.427
0.696
0.706
+0.181
-0.014 … 4.793
0.53
31
Sexual_Lust
strong
mlp2_h96_64
80,289
266,780
29,642
0.353
0.458
0.683
0.637
+0.250
-0.083 … 4.293
0.295
32
Malevolence_Malice
strong
mlp2_h64_32
51,457
266,781
29,641
0.386
0.463
0.679
0.466
+0.314
-0.042 … 3.603
0.263
33
Elation
strong
mlp2_h96_64
80,289
266,779
29,643
0.453
0.597
0.667
0.676
+0.277
-0.024 … 5.539
0.689
34
Valence
strong
mlp2_h96_64
80,289
266,780
29,642
0.588
0.841
0.645
0.624
-0.176
-4.555 … 4.106
0.572
35
Hope_Enthusiasm_Optimism
strong
mlp2_h64_32
51,457
266,780
29,642
0.598
0.777
0.64
0.611
+0.357
-0.014 … 6.184
0.93
36
Contempt
strong
mlp2_h64_32
51,457
266,779
29,643
0.41
0.486
0.638
0.519
+0.325
-0.013 … 4.734
0.269
37
Astonishment_Surprise
strong
mlp2_h64_32
51,457
266,781
29,641
0.516
0.661
0.632
0.561
+0.398
-0.029 … 5.133
0.446
38
Contentment
strong
mlp2_h64_32
51,457
266,779
29,643
0.309
0.404
0.63
0.655
+0.135
0.0 … 3.961
0.551
39
Bitterness
strong
mlp2_h96_64
80,289
266,781
29,641
0.344
0.402
0.612
0.54
+0.250
-0.032 … 4.254
0.204
40
Longing
strong
mlp2_h64_32
51,457
266,779
29,643
0.305
0.374
0.607
0.498
+0.233
-0.062 … 3.967
0.18
41
Doubt
medium
mlp2_h64_32
51,457
266,779
29,643
0.361
0.423
0.593
0.484
+0.271
-0.019 … 4.391
0.236
42
Confusion
medium
mlp2_h64_32
51,457
266,780
29,642
0.447
0.545
0.567
0.475
+0.354
-0.015 … 4.527
0.273
43
Affection
medium
mlp2_h64_32
51,457
266,779
29,643
0.511
0.639
0.55
0.465
+0.383
-0.023 … 5.215
0.399
44
Disgust
medium
mlp2_h64_32
51,457
266,779
29,643
0.399
0.459
0.543
0.39
+0.313
-0.026 … 3.543
0.216
45
Emotional_Numbness
medium
mlp2_h64_32
51,457
266,780
29,642
0.308
0.404
0.543
0.397
+0.242
-0.021 … 4.051
0.154
46
Sourness
medium
mlp2_h96_64
80,289
266,780
29,642
0.41
0.47
0.528
0.389
+0.316
-0.031 … 3.982
0.232
47
Triumph
medium
mlp2_h64_32
51,457
266,780
29,642
0.477
0.558
0.51
0.403
+0.381
-0.053 … 4.832
0.217
48
Sadness
medium
mlp2_h64_32
51,457
266,780
29,642
0.433
0.542
0.503
0.345
+0.372
-0.01 … 4.836
0.155
49
Pride
medium
mlp2_h96_64
80,289
266,778
29,644
0.397
0.49
0.502
0.522
+0.280
-0.009 … 4.629
0.298
50
Thankfulness_Gratitude
medium
mlp2_h96_64
80,289
266,780
29,642
0.762
0.916
0.495
0.411
+0.652
-0.056 … 4.832
0.425
51
Helplessness
medium
mlp2_h64_32
51,457
266,780
29,642
0.383
0.479
0.485
0.557
+0.321
-0.019 … 3.705
0.129
52
Distress
medium
mlp2_h64_32
51,457
266,781
29,641
0.501
0.626
0.481
0.289
+0.436
-0.0 … 4.699
0.159
53
Fatigue_Exhaustion
medium
mlp2_h64_32
51,457
266,780
29,642
0.326
0.441
0.478
0.374
+0.271
-0.092 … 5.379
0.104
54
Disappointment
medium
mlp2_h96_64
80,289
266,779
29,643
0.502
0.595
0.465
0.434
+0.393
-0.0 … 4.152
0.289
55
Infatuation
medium
mlp2_h64_32
51,457
266,779
29,643
0.455
0.553
0.44
0.365
+0.395
-0.059 … 4.676
0.128
56
Jealousy_&_Envy
medium
mlp2_h96_64
80,289
266,781
29,641
0.356
0.43
0.425
0.356
+0.293
-0.052 … 4.078
0.169
57
Shame
medium
mlp2_h64_32
51,457
266,781
29,641
0.408
0.48
0.408
0.36
+0.348
-0.023 … 5.066
0.131
58
Pain
medium
mlp2_h96_64
80,289
266,780
29,642
0.214
0.321
0.392
0.199
+0.178
-0.02 … 5.07
0.046
59
Fear
medium
mlp2_h64_32
51,457
266,780
29,642
0.405
0.464
0.39
0.255
+0.349
-0.052 … 4.137
0.094
60
Relief
medium
mlp1_h64
49,281
266,778
29,644
0.594
0.709
0.355
0.317
+0.491
-0.057 … 5.191
0.297
61
Awe
medium
mlp2_h64_32
51,457
266,780
29,642
0.461
0.534
0.338
0.235
+0.413
-0.026 … 4.731
0.087
(Rows with bias above +0.35 are the ones caveat 2 is about. Fear +0.349 and Confusion +0.354 sit just under.)
