The merged parent used in these specialist comparisons is now released as Humaneness Voice Large 1.6 Alpha: 1.5 + final RealSpeech rank 4, merged once. Apply only the selected new bucket adapter. Do not reload either ancestor.
1.5 Original model · 1.6 merged model · Complete context manuals for humans and agents
Read 1.5 context manual · Read 1.6 + adapter context manual. Exact prompts, CFG, adapter scales and failed gates are retained. New model/documentation contributions: CC BY 4.0; existing upstream and gated specialist terms remain applicable.
The original grid used generic GENERAL/SCRIPT directions at CFG1. It did not systematically use target-specific manual winners. The new study freezes the actual rank-4 manual wording, preserves every literal word, seed, reference, adapter and frame budget, then independently varies CFG and adapter strength. The previous recordings and conclusions remain below.
Ranking:100% requested EARS emotion/style head, standardized using frozen training statistics; low-pole targets reverse the sign. Formal/casual contrasts average the desired head and negative opposite head. CE and genuineness are diagnostics. Higher target z means stronger desired delivery, not a human rating. More CFG is not automatically better.
Selection:20 original-text outcomes per configuration select wording, CFG and scale. Fifty additional unseen-text reference pairs test the frozen choices. A recommended adapter must improve against both the same-CFG no-adapter control and the best manual-only setting. Quality gates: per-take WER≤15%, mean WER≤10%, valid/terminated fraction≥90%, paired identity change≥−0.05 over50pairs. Failed outcomes stay visible. Language-cluster bootstrap intervals use only five fixed reference voices; findings across72targets remain exploratory.
Five languages: EN/DE/FR/NL/ES; original seeds1701/1702; reference187frames/14.96seconds. CFG1/1.25/1.5/2/2.5/3/4/5 × bucket strength0/0.5/0.75/1. CFG4/5 are stress conditions, not presumed improvements. No-reference use of these bucket recommendations has not been validated.
Taxonomy:S_MONO means Monologue, S_TECH means Teacher/Didactic, VOLT means Volatility. The two corrected style recipes and17dimension/contrast recipes are new searches, not historical manual winners.53other targets reuse directly applicable manual recipes.
Best tested agent route: No new configuration passes the held-out gates; retain standard fallback and inspect audio. Bucket scale 0.0, CFG 1.0.
| Held-out condition | n | Target z↑ | Raw head | CE↑ | Genuine↑ | Voice↑ | WER↓ | Valid | QA |
|---|---|---|---|---|---|---|---|---|---|
| standard | 50 | 1.813 | 1.995 | 6.243 | 4.396 | 0.616 | 0.011 | 0.980 | pass |
| manual | 50 | 1.958 | 2.101 | 5.963 | 4.424 | 0.527 | 0.025 | 0.980 | fail |
| adapter | 50 | 1.873 | 2.039 | 6.208 | 4.395 | 0.574 | 0.008 | 1.000 | pass |
| matched_no_adapter | 50 | 1.880 | 2.044 | 6.181 | 4.391 | 0.609 | 0.007 | 1.000 | pass |
Incremental adapter gain versus same-CFG control: -0.006 z;95% cluster interval [-0.08468746754443514, 0.08234377619645952]. Versus best manual-only: -0.084 z. Manual-only versus standard: 0.145 z.
| Scale | CFG | Target z | CE | Genuine | WER | Provisional QA |
|---|---|---|---|---|---|---|
| 0.0 | 1.0 | 1.699 | 6.169 | 4.354 | 0.000 | True |
| 0.5 | 1.0 | 1.596 | 6.158 | 4.369 | 0.000 | True |
| 0.75 | 1.0 | 1.584 | 6.197 | 4.425 | 0.000 | True |
| 1.0 | 1.0 | 1.637 | 6.168 | 4.410 | 0.000 | True |
| 0.0 | 1.25 | 1.690 | 6.123 | 4.339 | 0.000 | True |
| 0.5 | 1.25 | 1.696 | 6.116 | 4.318 | 0.009 | True |
| 0.75 | 1.25 | 1.634 | 6.135 | 4.337 | 0.000 | True |
| 1.0 | 1.25 | 1.743 | 6.118 | 4.339 | 0.009 | True |
| 0.0 | 1.5 | 1.659 | 6.109 | 4.364 | 0.006 | True |
| 0.5 | 1.5 | 1.813 | 6.124 | 4.326 | 0.011 | True |
| 0.75 | 1.5 | 1.694 | 6.150 | 4.379 | 0.014 | True |
| 1.0 | 1.5 | 1.704 | 6.139 | 4.295 | 0.003 | True |
| 0.0 | 2.0 | 1.700 | 6.106 | 4.386 | 0.003 | True |
| 0.5 | 2.0 | 1.708 | 6.102 | 4.355 | 0.000 | True |
| 0.75 | 2.0 | 1.802 | 6.061 | 4.350 | 0.003 | True |
| 1.0 | 2.0 | 1.710 | 6.074 | 4.330 | 0.007 | True |
| 0.0 | 2.5 | 1.650 | 6.068 | 4.432 | 0.000 | True |
| 0.5 | 2.5 | 1.739 | 6.051 | 4.367 | 0.006 | True |
| 0.75 | 2.5 | 1.696 | 6.044 | 4.389 | 0.003 | True |
| 1.0 | 2.5 | 1.749 | 6.059 | 4.403 | 0.006 | True |
| 0.0 | 3.0 | 1.771 | 5.920 | 4.335 | 0.003 | True |
| 0.5 | 3.0 | 1.786 | 6.014 | 4.393 | 0.000 | True |
| 0.75 | 3.0 | 1.763 | 6.042 | 4.446 | 0.003 | True |
| 1.0 | 3.0 | 1.723 | 5.969 | 4.440 | 0.023 | True |
| 0.0 | 4.0 | 1.601 | 5.882 | 4.408 | 0.000 | True |
| 0.5 | 4.0 | 1.738 | 5.889 | 4.388 | 0.006 | True |
| 0.75 | 4.0 | 1.688 | 5.868 | 4.397 | 0.024 | True |
| 1.0 | 4.0 | 1.731 | 5.842 | 4.341 | 0.009 | True |
| 0.0 | 5.0 | 1.776 | 5.736 | 4.373 | 0.021 | True |
| 0.5 | 5.0 | 1.712 | 5.855 | 4.410 | 0.065 | False |
| 0.75 | 5.0 | 1.725 | 5.763 | 4.357 | 0.019 | True |
| 1.0 | 5.0 | 1.712 | 5.691 | 4.323 | 0.067 | False |
direct_manual_recipe
{
"id": "manual-04d8f738e9e1",
"general": "Employ a highly rhythmic and measured delivery, where each syllable is placed with the precision of a complex calculation.",
"inline": [
"Begin with a very low, grounded vocal placement, establishing a sense of unshakeable mental foundation.",
"Maintain near-perfect, deliberate pacing throughout, using slight pauses between conceptual units for emphasis.",
"Conclude with a brief, held vocal sustain on the final vowel, allowing the auditory impression of prolonged thought to linger."
],
"cfg": 2.5,
"recipe": "P07",
"strategy": "local-Gemma-mutation",
"timing_syntax": "canonical",
"canonical_labels": true,
"neutral_event": "keep",
"parent": "cfg-2.5",
"origin": "exact_frozen_manual_recipe"
}Exact machine-readable comparison · Agent tool recommendations
Intense Concentration. Base: 1.5 Alpha + merged real-speech rank4; new adapter: rank32. Every row is a matched seed/text/reference test. Scale zero disables the new adapter; all other scales are applied absolutely, never compounded.
Adapter training data: {"de": 3, "fr": 5, "en": 22, "nl": 1}. All available accepted languages were pooled, globally shuffled, for three epochs. Peak LR 2e-05 at the end of the first-epoch warmup, then cosine decay. All original optimizer/RNG checkpoints remain saved.
Taxonomy: S_MONO is monologue, not monotone; VOLT is volatility, not volume. No claim is made that a classifier score alone proves the audible performance.
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 20 | 0.941 ± 0.039 | 6.170 ± 0.161 | 4.435 ± 0.220 | 1.13% ± 3.00 |
| Rank32 × 0.5 | 20 | 0.932 ± 0.056 | 6.188 ± 0.124 | 4.407 ± 0.197 | 0.26% ± 1.18 |
| Rank32 × 0.75 | 20 | 0.931 ± 0.058 | 6.163 ± 0.154 | 4.431 ± 0.195 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 20 | 0.943 ± 0.056 | 6.164 ± 0.140 | 4.344 ± 0.184 | 0.00% ± 0.00 |
Fixed speaker reference:
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 4 | 0.908 ± 0.019 | 6.299 ± 0.106 | 4.274 ± 0.278 | 0.00% ± 0.00 |
| Rank32 × 0.5 | 4 | 0.859 ± 0.059 | 6.279 ± 0.091 | 4.307 ± 0.202 | 0.00% ± 0.00 |
| Rank32 × 0.75 | 4 | 0.855 ± 0.064 | 6.324 ± 0.122 | 4.310 ± 0.282 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 4 | 0.891 ± 0.061 | 6.279 ± 0.072 | 4.168 ± 0.100 | 0.00% ± 0.00 |
| Prompt / seed | Merged base · no new LoRA | Rank32 × 0.5 | Rank32 × 0.75 | Rank32 × 1.0 |
|---|---|---|---|---|
| Seed 1701 The door is still open. I put your coat on the chair. Tell me when you are ready. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] The door is still open. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] I put your coat on the chair. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Tell me when you are ready. | Target 0.936 · CE 6.28 · Genuine 4.00 · WER 0.0% | Target 0.803 · CE 6.22 · Genuine 4.28 · WER 0.0% | Target 0.760 · CE 6.25 · Genuine 4.17 · WER 0.0% | Target 0.914 · CE 6.33 · Genuine 4.14 · WER 0.0% |
| Seed 1702 The door is still open. I put your coat on the chair. Tell me when you are ready. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] The door is still open. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] I put your coat on the chair. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Tell me when you are ready. | Target 0.897 · CE 6.16 · Genuine 4.18 · WER 0.0% | Target 0.896 · CE 6.18 · Genuine 4.03 · WER 0.0% | Target 0.899 · CE 6.21 · Genuine 3.99 · WER 0.0% | Target 0.941 · CE 6.17 · Genuine 4.06 · WER 0.0% |
| Seed 1701 I found the little box. It was beside the old window. We can look inside together. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] I found the little box. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] It was beside the old window. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] We can look inside together. | Target 0.894 · CE 6.39 · Genuine 4.66 · WER 0.0% | Target 0.922 · CE 6.35 · Genuine 4.46 · WER 0.0% | Target 0.876 · CE 6.48 · Genuine 4.62 · WER 0.0% | Target 0.803 · CE 6.29 · Genuine 4.19 · WER 0.0% |
| Seed 1702 I found the little box. It was beside the old window. We can look inside together. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] I found the little box. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] It was beside the old window. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] We can look inside together. | Target 0.905 · CE 6.37 · Genuine 4.26 · WER 0.0% | Target 0.813 · CE 6.36 · Genuine 4.46 · WER 0.0% | Target 0.886 · CE 6.36 · Genuine 4.46 · WER 0.0% | Target 0.907 · CE 6.32 · Genuine 4.29 · WER 0.0% |
Fixed speaker reference:
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 4 | 0.909 ± 0.059 | 6.332 ± 0.096 | 4.520 ± 0.254 | 1.32% ± 2.63 |
| Rank32 × 0.5 | 4 | 0.913 ± 0.052 | 6.311 ± 0.104 | 4.451 ± 0.116 | 0.00% ± 0.00 |
| Rank32 × 0.75 | 4 | 0.912 ± 0.047 | 6.281 ± 0.151 | 4.438 ± 0.144 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 4 | 0.914 ± 0.085 | 6.271 ± 0.089 | 4.459 ± 0.203 | 0.00% ± 0.00 |
| Prompt / seed | Merged base · no new LoRA | Rank32 × 0.5 | Rank32 × 0.75 | Rank32 × 1.0 |
|---|---|---|---|---|
| Seed 1701 Die Tür ist noch offen. Ich habe deinen Mantel auf den Stuhl gelegt. Sag mir, wenn du bereit bist. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Die Tür ist noch offen. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Ich habe deinen Mantel auf den Stuhl gelegt. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Sag mir, wenn du bereit bist. | Target 0.871 · CE 6.21 · Genuine 4.68 · WER 5.3% | Target 0.838 · CE 6.26 · Genuine 4.61 · WER 0.0% | Target 0.886 · CE 6.30 · Genuine 4.48 · WER 0.0% | Target 0.792 · CE 6.17 · Genuine 4.73 · WER 0.0% |
| Seed 1702 Die Tür ist noch offen. Ich habe deinen Mantel auf den Stuhl gelegt. Sag mir, wenn du bereit bist. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Die Tür ist noch offen. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Ich habe deinen Mantel auf den Stuhl gelegt. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Sag mir, wenn du bereit bist. | Target 0.845 · CE 6.32 · Genuine 4.77 · WER 0.0% | Target 0.930 · CE 6.19 · Genuine 4.47 · WER 0.0% | Target 0.863 · CE 6.06 · Genuine 4.29 · WER 0.0% | Target 0.924 · CE 6.23 · Genuine 4.41 · WER 0.0% |
| Seed 1701 Ich habe die kleine Schachtel gefunden. Sie stand neben dem alten Fenster. Wir können gemeinsam hineinschauen. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Ich habe die kleine Schachtel gefunden. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Sie stand neben dem alten Fenster. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Wir können gemeinsam hineinschauen. | Target 0.959 · CE 6.36 · Genuine 4.22 · WER 0.0% | Target 0.959 · CE 6.38 · Genuine 4.37 · WER 0.0% | Target 0.966 · CE 6.36 · Genuine 4.36 · WER 0.0% | Target 0.969 · CE 6.38 · Genuine 4.44 · WER 0.0% |
| Seed 1702 Ich habe die kleine Schachtel gefunden. Sie stand neben dem alten Fenster. Wir können gemeinsam hineinschauen. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Ich habe die kleine Schachtel gefunden. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Sie stand neben dem alten Fenster. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Wir können gemeinsam hineinschauen. | Target 0.960 · CE 6.44 · Genuine 4.41 · WER 0.0% | Target 0.922 · CE 6.41 · Genuine 4.36 · WER 0.0% | Target 0.935 · CE 6.40 · Genuine 4.62 · WER 0.0% | Target 0.974 · CE 6.30 · Genuine 4.24 · WER 0.0% |
Fixed speaker reference:
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 4 | 0.941 ± 0.008 | 6.086 ± 0.164 | 4.536 ± 0.132 | 1.39% ± 2.78 |
| Rank32 × 0.5 | 4 | 0.947 ± 0.031 | 6.162 ± 0.056 | 4.558 ± 0.096 | 0.00% ± 0.00 |
| Rank32 × 0.75 | 4 | 0.941 ± 0.005 | 6.126 ± 0.087 | 4.467 ± 0.097 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 4 | 0.963 ± 0.010 | 6.185 ± 0.057 | 4.454 ± 0.113 | 0.00% ± 0.00 |
| Prompt / seed | Merged base · no new LoRA | Rank32 × 0.5 | Rank32 × 0.75 | Rank32 × 1.0 |
|---|---|---|---|---|
| Seed 1701 La porte est encore ouverte. J’ai posé ton manteau sur la chaise. Dis-moi quand tu es prêt. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] La porte est encore ouverte. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] J’ai posé ton manteau sur la chaise. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Dis-moi quand tu es prêt. | Target 0.935 · CE 6.05 · Genuine 4.46 · WER 0.0% | Target 0.961 · CE 6.13 · Genuine 4.70 · WER 0.0% | Target 0.945 · CE 6.20 · Genuine 4.51 · WER 0.0% | Target 0.971 · CE 6.12 · Genuine 4.31 · WER 0.0% |
| Seed 1702 La porte est encore ouverte. J’ai posé ton manteau sur la chaise. Dis-moi quand tu es prêt. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] La porte est encore ouverte. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] J’ai posé ton manteau sur la chaise. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Dis-moi quand tu es prêt. | Target 0.949 · CE 5.87 · Genuine 4.48 · WER 5.6% | Target 0.957 · CE 6.13 · Genuine 4.55 · WER 0.0% | Target 0.937 · CE 6.01 · Genuine 4.37 · WER 0.0% | Target 0.957 · CE 6.24 · Genuine 4.55 · WER 0.0% |
| Seed 1701 J’ai trouvé la petite boîte. Elle était près de la vieille fenêtre. Nous pouvons regarder ensemble ce qu’elle contient. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] J’ai trouvé la petite boîte. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Elle était près de la vieille fenêtre. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Nous pouvons regarder ensemble ce qu’elle contient. | Target 0.933 · CE 6.20 · Genuine 4.73 · WER 0.0% | Target 0.969 · CE 6.14 · Genuine 4.48 · WER 0.0% | Target 0.946 · CE 6.12 · Genuine 4.58 · WER 0.0% | Target 0.971 · CE 6.23 · Genuine 4.55 · WER 0.0% |
| Seed 1702 J’ai trouvé la petite boîte. Elle était près de la vieille fenêtre. Nous pouvons regarder ensemble ce qu’elle contient. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] J’ai trouvé la petite boîte. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Elle était près de la vieille fenêtre. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Nous pouvons regarder ensemble ce qu’elle contient. | Target 0.945 · CE 6.23 · Genuine 4.47 · WER 0.0% | Target 0.901 · CE 6.25 · Genuine 4.51 · WER 0.0% | Target 0.937 · CE 6.18 · Genuine 4.41 · WER 0.0% | Target 0.951 · CE 6.16 · Genuine 4.41 · WER 0.0% |
Fixed speaker reference:
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 4 | 0.972 ± 0.012 | 6.134 ± 0.062 | 4.368 ± 0.084 | 0.00% ± 0.00 |
| Rank32 × 0.5 | 4 | 0.972 ± 0.010 | 6.134 ± 0.036 | 4.371 ± 0.076 | 1.32% ± 2.63 |
| Rank32 × 0.75 | 4 | 0.985 ± 0.014 | 6.031 ± 0.093 | 4.344 ± 0.135 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 4 | 0.978 ± 0.028 | 6.086 ± 0.142 | 4.289 ± 0.080 | 0.00% ± 0.00 |
| Prompt / seed | Merged base · no new LoRA | Rank32 × 0.5 | Rank32 × 0.75 | Rank32 × 1.0 |
|---|---|---|---|---|
| Seed 1701 De deur staat nog open. Ik heb je jas op de stoel gelegd. Zeg maar wanneer je klaar bent. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] De deur staat nog open. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Ik heb je jas op de stoel gelegd. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Zeg maar wanneer je klaar bent. | Target 0.980 · CE 6.13 · Genuine 4.34 · WER 0.0% | Target 0.981 · CE 6.17 · Genuine 4.36 · WER 0.0% | Target 0.993 · CE 5.93 · Genuine 4.15 · WER 0.0% | Target 0.993 · CE 6.08 · Genuine 4.24 · WER 0.0% |
| Seed 1702 De deur staat nog open. Ik heb je jas op de stoel gelegd. Zeg maar wanneer je klaar bent. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] De deur staat nog open. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Ik heb je jas op de stoel gelegd. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Zeg maar wanneer je klaar bent. | Target 0.976 · CE 6.14 · Genuine 4.26 · WER 0.0% | Target 0.974 · CE 6.14 · Genuine 4.48 · WER 5.3% | Target 0.990 · CE 6.13 · Genuine 4.46 · WER 0.0% | Target 0.991 · CE 6.19 · Genuine 4.36 · WER 0.0% |
| Seed 1701 Ik heb het kleine doosje gevonden. Het stond naast het oude raam. We kunnen er samen in kijken. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Ik heb het kleine doosje gevonden. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Het stond naast het oude raam. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] We kunnen er samen in kijken. | Target 0.978 · CE 6.05 · Genuine 4.46 · WER 0.0% | Target 0.976 · CE 6.13 · Genuine 4.33 · WER 0.0% | Target 0.994 · CE 5.98 · Genuine 4.40 · WER 0.0% | Target 0.935 · CE 6.19 · Genuine 4.36 · WER 0.0% |
| Seed 1702 Ik heb het kleine doosje gevonden. Het stond naast het oude raam. We kunnen er samen in kijken. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Ik heb het kleine doosje gevonden. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Het stond naast het oude raam. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] We kunnen er samen in kijken. | Target 0.955 · CE 6.21 · Genuine 4.41 · WER 0.0% | Target 0.958 · CE 6.09 · Genuine 4.32 · WER 0.0% | Target 0.964 · CE 6.08 · Genuine 4.37 · WER 0.0% | Target 0.992 · CE 5.89 · Genuine 4.20 · WER 0.0% |
No accepted bucket-training example in this language: transfer evaluation only.
Fixed speaker reference:
| Condition | n | Target score ↑ | Content Enjoyment ↑ | Genuineness ↑ | WER ↓ |
|---|---|---|---|---|---|
| Merged base, no bucket adapter | 4 | 0.973 ± 0.006 | 5.997 ± 0.069 | 4.479 ± 0.271 | 2.94% ± 5.88 |
| Rank32 × 0.5 | 4 | 0.970 ± 0.017 | 6.054 ± 0.118 | 4.350 ± 0.354 | 0.00% ± 0.00 |
| Rank32 × 0.75 | 4 | 0.963 ± 0.034 | 6.052 ± 0.048 | 4.597 ± 0.213 | 0.00% ± 0.00 |
| Rank32 × 1.0 | 4 | 0.969 ± 0.017 | 5.999 ± 0.093 | 4.352 ± 0.256 | 0.00% ± 0.00 |
| Prompt / seed | Merged base · no new LoRA | Rank32 × 0.5 | Rank32 × 0.75 | Rank32 × 1.0 |
|---|---|---|---|---|
| Seed 1701 La puerta sigue abierta. He dejado tu abrigo sobre la silla. Dime cuándo estás listo. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] La puerta sigue abierta. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] He dejado tu abrigo sobre la silla. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Dime cuándo estás listo. | Target 0.978 · CE 5.97 · Genuine 4.17 · WER 0.0% | Target 0.976 · CE 6.17 · Genuine 4.32 · WER 0.0% | Target 0.976 · CE 6.00 · Genuine 4.36 · WER 0.0% | Target 0.945 · CE 6.08 · Genuine 4.31 · WER 0.0% |
| Seed 1702 La puerta sigue abierta. He dejado tu abrigo sobre la silla. Dime cuándo estás listo. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] La puerta sigue abierta. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] He dejado tu abrigo sobre la silla. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Dime cuándo estás listo. | Target 0.968 · CE 5.94 · Genuine 4.38 · WER 0.0% | Target 0.982 · CE 6.02 · Genuine 3.89 · WER 0.0% | Target 0.912 · CE 6.07 · Genuine 4.48 · WER 0.0% | Target 0.984 · CE 5.86 · Genuine 4.02 · WER 0.0% |
| Seed 1701 He encontrado la cajita. Estaba junto a la ventana vieja. Podemos mirar juntos lo que hay dentro. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] He encontrado la cajita. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Estaba junto a la ventana vieja. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Podemos mirar juntos lo que hay dentro. | Target 0.977 · CE 5.99 · Genuine 4.81 · WER 0.0% | Target 0.976 · CE 5.90 · Genuine 4.44 · WER 0.0% | Target 0.980 · CE 6.11 · Genuine 4.83 · WER 0.0% | Target 0.976 · CE 6.02 · Genuine 4.46 · WER 0.0% |
| Seed 1702 He encontrado la cajita. Estaba junto a la ventana vieja. Podemos mirar juntos lo que hay dentro. Exact GENERAL / SCRIPT promptGENERAL: One adult speaker in a clean recording. Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word. SCRIPT: (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] He encontrado la cajita. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.8 seconds duration] Estaba junto a la ventana vieja. (Intense Concentration. Make the requested delivery clearly audible, not merely implied by the words. Preserve every word.) [3.2 seconds duration] Podemos mirar juntos lo que hay dentro. | Target 0.967 · CE 6.09 · Genuine 4.56 · WER 11.8% | Target 0.945 · CE 6.12 · Genuine 4.75 · WER 0.0% | Target 0.983 · CE 6.03 · Genuine 4.71 · WER 0.0% | Target 0.972 · CE 6.03 · Genuine 4.62 · WER 0.0% |