Podcast rank-64 LoRA: two starting checkpoints

Compare the original NotebookLM full fine-tune, that model plus a new rank-64 LoRA, Acting High-8654, and that model plus its new rank-64 LoRA. All 50 frozen held-out tasks use identical two-speaker references, earlier conversation, prompt, seed and frame budget. Five two-turn and five four-turn tasks per language. Each output is generated jointly, not stitched. Sampling: audio temperature 1, top-p .95, top-k 50, CFG 1; one take.

The adapters were trained on 732 unique clean examples across EN/DE/FR/ES, with six nested quality stages and approximately 2.97 hours of presented target audio including repetitions. Dutch was excluded from adapter training and is a transfer test; its publisher language labels were reported by the user as unreliable. Original train/test episode separation is preserved. Scores are EARS/ASR predictions, not listener ratings. Entire outputs are scored in duration-weighted windows of at most 30 s; WER is the mean per-clip error rate and concatenates nonoverlapping ASR windows, so chunk-boundary errors are possible. Source top-two emotion MSE uses fixed heads from the held-out source continuation, not the identity references. Speaker-switch accuracy and speaker identity are listening questions here; they are not falsely inferred from requested script timings.

All languages

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base506.0843.9437.0943.0240.0780.0220.0010
NotebookLM + rank 64 · final 652506.0793.9597.0953.0040.1140.3300.0010
Acting High 8,654 · base506.2363.9277.2702.9460.0810.1660.0040
Acting High 8,654 + rank 64 · final 652506.1653.9677.0222.9400.1380.1280.0020

EN

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base106.1874.0476.7442.9910.1170.0080.0000
NotebookLM + rank 64 · final 652106.2134.0126.7212.9730.1030.0240.0010
Acting High 8,654 · base106.2813.6826.5192.9430.0970.1680.0010
Acting High 8,654 + rank 64 · final 652106.2543.8476.5722.9590.1180.1920.0010

DE

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base106.1683.7446.6133.1000.0880.0080.0010
NotebookLM + rank 64 · final 652106.1313.7686.5443.1130.1471.5360.0020
Acting High 8,654 · base106.3213.9756.9663.0680.0560.1360.0090
Acting High 8,654 + rank 64 · final 652106.2843.9106.5152.8350.1390.0800.0010

FR

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base106.1364.1197.4262.8730.0320.0320.0020
NotebookLM + rank 64 · final 652106.1344.1587.5012.7690.1760.0160.0000
Acting High 8,654 · base106.2524.0397.6062.6910.1150.1760.0010
Acting High 8,654 + rank 64 · final 652106.2244.1627.3632.8510.0610.1360.0030

ES

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base105.9394.0377.5722.9350.1350.0240.0010
NotebookLM + rank 64 · final 652105.9544.0467.6402.8960.0930.0160.0020
Acting High 8,654 · base106.1794.1458.1272.7300.1210.2960.0020
Acting High 8,654 + rank 64 · final 652106.0214.0607.6612.7170.3370.1200.0030

NL · transfer

ConditionNCE ↑Genuineness ↑Blend ↑ArousalWER ↓Duration error s ↓Source top-2 MSE ↓Frame caps
NotebookLM full FT · base105.9903.7697.1133.2230.0160.0400.0020
NotebookLM + rank 64 · final 652105.9633.8087.0713.2690.0520.0560.0020
Acting High 8,654 · base106.1493.7967.1323.2960.0180.0560.0040
Acting High 8,654 + rank 64 · final 652106.0453.8577.0003.3390.0330.1120.0030

Matched listening

en-2turn-00

Exact text and prompt

Sounds complex, but amp handles it. Mostly, yeah. You just need to wrap your forward pass in the autocast context and use the grad scalar around your optimizer step.

GENERAL: 
SCRIPT:
(speaker=S1, understated, steady, enjoyable) [2.04 seconds duration] Sounds complex, but amp handles it.
(speaker=S2, very choppy, very even-tempered, lightness, enjoyable) [0.96 seconds duration] Mostly, yeah.
(speaker=S2, very news-anchor, very even-rhythm, tranquility, enjoyable) [5.88 seconds duration] You just need to wrap your forward pass in the autocast context and use the grad scalar around your optimizer step.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.18 · genuine 4.14 · WER 0.0%
8.88 s; requested 8.88 s

CE 6.21 · genuine 3.99 · WER 0.0%
8.88 s; requested 8.88 s

CE 6.10 · genuine 3.76 · WER 0.0%
8.96 s; requested 8.88 s

CE 6.30 · genuine 3.72 · WER 0.0%
8.96 s; requested 8.88 s

en-2turn-01

Exact text and prompt

torches stot distributed. When one GPU isn't enough, you need to train across multiple GPUs, potentially across multiple machines. How does that work broadly?

GENERAL: 
SCRIPT:
(speaker=S2, very clean-bg, steady, detachment, enjoyable) [1.64 seconds duration] torches stot distributed.
(speaker=S2, very news-anchor, very hi-fi, concentration, enjoyable) [6.00 seconds duration] When one GPU isn't enough, you need to train across multiple GPUs, potentially across multiple machines.
(speaker=S1, very clean-bg, steady, enjoyable) [1.40 seconds duration] How does that work broadly?
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.34 · genuine 3.89 · WER 4.2%
9.04 s; requested 9.04 s

CE 6.50 · genuine 3.90 · WER 4.2%
9.04 s; requested 9.04 s

CE 6.40 · genuine 3.78 · WER 4.2%
9.12 s; requested 9.04 s

CE 6.36 · genuine 3.54 · WER 0.0%
9.04 s; requested 9.04 s

en-2turn-02

Exact text and prompt

compared to standard 32-bit FP32. So faster training.

GENERAL: 
SCRIPT:
(speaker=S2, even-tempered, news-anchor, enjoyable) [2.40 seconds duration] compared to standard 32-bit FP32.
(speaker=S1, very accelerating, very clean-bg, slightly honor, enjoyable) [1.04 seconds duration] So faster training.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.18 · genuine 4.22 · WER 55.6%
3.44 s; requested 3.44 s

CE 6.15 · genuine 4.13 · WER 55.6%
3.44 s; requested 3.44 s

CE 6.10 · genuine 3.69 · WER 66.7%
3.84 s; requested 3.44 s

CE 6.11 · genuine 4.53 · WER 55.6%
3.52 s; requested 3.44 s

en-2turn-03

Exact text and prompt

takes your dataset object and handles several vital things automatically. Like

GENERAL: 
SCRIPT:
(speaker=S2, very long-phrased, very news-anchor, slightly engrossment, enjoyable) [3.84 seconds duration] takes your dataset object and handles several vital things automatically.
(speaker=S1, very flat, very cold, slightly boredom, less-enjoyable) [0.16 seconds duration] Like
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.92 · genuine 4.51 · WER 18.2%
4.00 s; requested 4.00 s

CE 5.94 · genuine 4.36 · WER 18.2%
4.00 s; requested 4.00 s

CE 6.21 · genuine 4.53 · WER 0.0%
4.24 s; requested 4.00 s

CE 6.25 · genuine 4.37 · WER 9.1%
4.00 s; requested 4.00 s

en-2turn-04

Exact text and prompt

TensorFlow static graph compilation at tf dot function can sometimes yield better optimization opportunities for stable production models, but torch.compile again?

GENERAL: 
SCRIPT:
(speaker=S2, very news-anchor, very even-tempered, calmness, enjoyable) [8.32 seconds duration] TensorFlow static graph compilation at tf dot function can sometimes yield better optimization opportunities for stable production models, but
[0.32 seconds pause]
(speaker=S1, very non-narrative, very accelerating, slightly ease, enjoyable) [1.20 seconds duration] torch.compile again?
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.34 · genuine 4.00 · WER 13.6%
9.76 s; requested 9.84 s

CE 6.33 · genuine 3.87 · WER 9.1%
9.76 s; requested 9.84 s

CE 6.32 · genuine 3.80 · WER 9.1%
9.84 s; requested 9.84 s

CE 6.23 · genuine 3.79 · WER 9.1%
9.76 s; requested 9.84 s

en-4turn-00

Exact text and prompt

Batching is key. Okay. Lots of good detective work and optimization strategies there. Twitch.combo sounds particularly exciting. It really is a major step forward for PyTorch performance.

GENERAL: 
SCRIPT:
(speaker=S1, very accelerating, very dark, slightly self-consciousness) [0.96 seconds duration] Batching is key.
(speaker=S2, very non-narrative, very non-narrative, respite, less-enjoyable) [0.40 seconds duration] Okay.
(speaker=S1, very news-anchor, very full, slightly relaxation, enjoyable) [2.80 seconds duration] Lots of good detective work and optimization strategies there.
(speaker=S1, very even-tempered, very warm, courage, enjoyable) [1.04 seconds duration] Twitch.combo sounds
[0.32 seconds pause]
(speaker=S1, very even-tempered, very warm, courage, enjoyable) [0.96 seconds duration] particularly exciting.
(speaker=S2, very news-anchor, very clean-bg, slightly fulfillment, enjoyable) [2.56 seconds duration] It really is a major step forward for PyTorch performance.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.30 · genuine 4.18 · WER 3.6%
9.04 s; requested 9.04 s

CE 6.31 · genuine 4.25 · WER 3.6%
9.12 s; requested 9.04 s

CE 6.24 · genuine 3.55 · WER 0.0%
9.36 s; requested 9.04 s

CE 6.17 · genuine 3.85 · WER 14.3%
9.44 s; requested 9.04 s

en-4turn-01

Exact text and prompt

He actually criticized all three sex. Really? Yeah, he saw strengths and weaknesses in each. What did he think of the empiricists, for example?

GENERAL: 
SCRIPT:
(speaker=S2, very news-anchor, very tonal) [2.40 seconds duration] He actually criticized all three sex.
(speaker=S1, very accelerating, very non-narrative, astonishment) [0.32 seconds duration] Really?
(speaker=S2, very tonal, even-rhythm, slightly interest, enjoyable) [2.80 seconds duration] Yeah, he saw strengths and weaknesses in each.
(speaker=S1, chesty, full, uncertainty, enjoyable) [2.64 seconds duration] What did he think of the empiricists, for example?
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.01 · genuine 4.26 · WER 4.2%
8.16 s; requested 8.16 s

CE 6.03 · genuine 4.13 · WER 4.2%
8.16 s; requested 8.16 s

CE 6.39 · genuine 3.76 · WER 4.2%
8.32 s; requested 8.16 s

CE 6.21 · genuine 4.03 · WER 4.2%
8.32 s; requested 8.16 s

en-4turn-02

Exact text and prompt

in-place operations cautiously where appropriate. Batch operations whenever possible, operating on a large batch is usually much faster than operating on individual items sequentially. Batching is key. Okay. Lots of good detective work and optimization strategies there. Twitch.combo sounds particularly exciting.

GENERAL: 
SCRIPT:
(speaker=S2, very news-anchor, smooth, slightly deep focus, enjoyable) [2.64 seconds duration] in-place operations cautiously where appropriate.
(speaker=S2, very news-anchor, very long-phrased, slightly deep focus, enjoyable) [6.24 seconds duration] Batch operations whenever possible, operating on a large batch is usually much faster than operating on individual items sequentially.
(speaker=S1, very accelerating, very dark, slightly self-consciousness) [1.04 seconds duration] Batching is key.
(speaker=S2, very non-narrative, very non-narrative, respite, less-enjoyable) [0.40 seconds duration] Okay.
(speaker=S1, very news-anchor, very full, slightly relaxation, enjoyable) [2.80 seconds duration] Lots of good detective work and optimization strategies there.
(speaker=S1, very even-tempered, very warm, courage, enjoyable) [1.04 seconds duration] Twitch.combo sounds
[0.32 seconds pause]
(speaker=S1, very even-tempered, very warm, courage, enjoyable) [0.96 seconds duration] particularly exciting.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.15 · genuine 3.82 · WER 2.3%
15.44 s; requested 15.44 s

CE 6.13 · genuine 3.83 · WER 0.0%
15.44 s; requested 15.44 s

CE 6.25 · genuine 3.33 · WER 0.0%
15.52 s; requested 15.44 s

CE 6.20 · genuine 3.29 · WER 4.7%
16.00 s; requested 15.44 s

en-4turn-03

Exact text and prompt

how does this idea of balance play out in Galen's ethics? What does it actually mean to live a balanced life? So for Galen, achieving this balance involves a couple of things. First, we have to train and discipline those non-rational parts of the soul, you know. the spirited and desiderative. Right.

GENERAL: 
SCRIPT:
(speaker=S1, warm, long-phrased, very curiosity, enjoyable) [3.08 seconds duration] how does this idea of balance play out in Galen's ethics?
(speaker=S1, very warm, very darkening, very brooding, enjoyable) [2.24 seconds duration] What does it actually mean to live a balanced life?
(speaker=S2, very news-anchor, very bright, serenity, enjoyable) [3.92 seconds duration] So for Galen, achieving this balance involves a couple of things.
(speaker=S2, very news-anchor, seamless-breath, slightly engrossment, enjoyable) [4.08 seconds duration] First, we have to train and discipline those non-rational parts of the soul, you know.
[0.48 seconds pause]
(speaker=S1, very full, very steady, very awe, enjoyable) [1.44 seconds duration] the spirited and desiderative.
(speaker=S2, very non-narrative, very flat, less-enjoyable) [0.44 seconds duration] Right.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.28 · genuine 4.10 · WER 9.4%
15.68 s; requested 15.68 s

CE 6.27 · genuine 4.32 · WER 5.7%
15.68 s; requested 15.68 s

CE 6.52 · genuine 3.74 · WER 3.8%
15.68 s; requested 15.68 s

CE 6.39 · genuine 3.75 · WER 5.7%
15.68 s; requested 15.68 s

en-4turn-04

Exact text and prompt

can then output summaries, like a table showing time spent in different operations, or export detailed traces. Traces for. Viewing in tools like Tensor Board's Profiler plugin or Chrome's Chrome.tracing, these give you a timeline view showing exactly when operations ran on the CPU and GPU, helps you spot gaps where the GPU is idle, often waiting for data. So look for GPU idle time, any other profilers.

GENERAL: 
SCRIPT:
(speaker=S2, very clean-bg, very news-anchor, serenity, enjoyable) [5.84 seconds duration] can then output summaries, like a table showing time spent in different operations, or export detailed traces.
(speaker=S1, very steady, very flat, apathy) [0.56 seconds duration] Traces for.
(speaker=S2, very news-anchor, very tonal, curiosity, enjoyable) [12.16 seconds duration] Viewing in tools like Tensor Board's Profiler plugin or Chrome's Chrome.tracing, these give you a timeline view showing exactly when operations ran on the CPU and GPU, helps you spot gaps where the GPU is idle, often waiting for
(speaker=S2, very news-anchor, very tonal, curiosity, enjoyable) [0.48 seconds duration] data.
(speaker=S1, very seamless-breath, very full, slightly peacefulness, enjoyable) [3.04 seconds duration] So look for GPU idle time, any other profilers.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.17 · genuine 3.35 · WER 5.8%
22.08 s; requested 22.08 s

CE 6.28 · genuine 3.33 · WER 2.9%
22.16 s; requested 22.08 s

CE 6.30 · genuine 2.88 · WER 8.7%
22.40 s; requested 22.08 s

CE 6.32 · genuine 3.60 · WER 15.9%
22.64 s; requested 22.08 s

de-2turn-00

Exact text and prompt

Genau. Wo Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt.

GENERAL: 
SCRIPT:
(speaker=S2, very non-narrative, very clean-bg) [0.44 seconds duration] Genau.
(speaker=S2, very non-narrative, very scattered, indifference, less-enjoyable) [0.08 seconds duration] Wo
(speaker=S1, very full, very news-anchor, concentration, enjoyable) [8.36 seconds duration] Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.25 · genuine 3.98 · WER 12.5%
8.88 s; requested 8.88 s

CE 6.27 · genuine 4.09 · WER 8.3%
8.88 s; requested 8.88 s

CE 6.32 · genuine 4.26 · WER 12.5%
8.96 s; requested 8.88 s

CE 6.25 · genuine 4.04 · WER 100.0%
8.88 s; requested 8.88 s

de-2turn-01

Exact text and prompt

Genau die Polizei wird da faktisch zum Souverän. Aber, wow, halt stopp, da bin ich beim Lesen innerlich echt auf die Barrikaden gegangen.

GENERAL: 
SCRIPT:
(speaker=S2, very formal, very news-anchor, slightly concentration, enjoyable) [3.76 seconds duration] Genau die Polizei wird da faktisch zum Souverän.
(speaker=S1, very full, very warm, shock, enjoyable) [4.88 seconds duration] Aber, wow, halt stopp, da bin ich beim Lesen innerlich echt auf die Barrikaden gegangen.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.15 · genuine 3.98 · WER 4.3%
8.64 s; requested 8.64 s

CE 6.15 · genuine 3.92 · WER 0.0%
8.64 s; requested 8.64 s

CE 6.28 · genuine 4.02 · WER 0.0%
8.64 s; requested 8.64 s

CE 6.32 · genuine 3.94 · WER 0.0%
8.64 s; requested 8.64 s

de-2turn-02

Exact text and prompt

Und hier wird Scharmas überraschend optimistisch. Er formuliert das Prinzip der organisationalen Invarianz. Was besagt das?

GENERAL: 
SCRIPT:
(speaker=S2, very bright, very oral-bright, very hope, enjoyable) [0.88 seconds duration] Und hier wird Scharmas
[0.24 seconds pause]
(speaker=S2, very bright, very oral-bright, very hope, enjoyable) [1.60 seconds duration] überraschend optimistisch.
(speaker=S2, very news-anchor, very bright, concentration, enjoyable) [3.36 seconds duration] Er formuliert das Prinzip der organisationalen
[0.24 seconds pause]
(speaker=S2, very news-anchor, very bright, concentration, enjoyable) [1.04 seconds duration] Invarianz.
(speaker=S1, very accelerating, clean-bg, pessimism) [0.88 seconds duration] Was besagt das?
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.05 · genuine 3.87 · WER 6.2%
8.24 s; requested 8.24 s

CE 6.05 · genuine 3.94 · WER 6.2%
8.32 s; requested 8.24 s

CE 6.42 · genuine 4.17 · WER 25.0%
8.72 s; requested 8.24 s

CE 6.28 · genuine 4.17 · WER 6.2%
8.16 s; requested 8.24 s

de-2turn-03

Exact text and prompt

Jeder auf der Straße darf ihn erschlagen. Und Es ist juristisch komplett irrelevant. Gleichzeitig ist er aber nicht einmal für die Götter wertvoll genug, um als Opfertier zu dienen.

GENERAL: 
SCRIPT:
(speaker=S2, very clean-content, very seamless-breath) [2.00 seconds duration] Jeder auf der Straße darf ihn erschlagen.
(speaker=S2, bright, thin-throat, detestation, enjoyable) [0.16 seconds duration] Und
[0.64 seconds pause]
(speaker=S2, bright, thin-throat, detestation, enjoyable) [2.72 seconds duration] Es ist juristisch komplett irrelevant.
(speaker=S1, very full, very blended-resonance, very gratitude, enjoyable) [4.48 seconds duration] Gleichzeitig ist er aber nicht einmal für die Götter wertvoll genug, um als Opfertier zu dienen.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.03 · genuine 3.70 · WER 10.3%
10.00 s; requested 10.00 s

CE 6.03 · genuine 3.80 · WER 0.0%
10.00 s; requested 10.00 s

CE 6.33 · genuine 3.94 · WER 0.0%
10.08 s; requested 10.00 s

CE 6.39 · genuine 4.03 · WER 3.4%
10.00 s; requested 10.00 s

de-2turn-04

Exact text and prompt

Molekül für Molekül Eine exakte physikalische Kopie. Okay.

GENERAL: 
SCRIPT:
(speaker=S2, very oral-bright, very forward-mask, deep focus, enjoyable) [1.44 seconds duration] Molekül für Molekül
[0.44 seconds pause]
(speaker=S2, very oral-bright, very forward-mask, deep focus, enjoyable) [2.32 seconds duration] Eine exakte physikalische Kopie.
(speaker=S1, very flat, very non-narrative, less-enjoyable) [0.44 seconds duration] Okay.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.12 · genuine 3.96 · WER 0.0%
4.64 s; requested 4.64 s

CE 5.97 · genuine 3.85 · WER 0.0%
4.64 s; requested 4.64 s

CE 6.27 · genuine 4.11 · WER 0.0%
4.96 s; requested 4.64 s

CE 6.06 · genuine 4.03 · WER 0.0%
4.72 s; requested 4.64 s

de-4turn-00

Exact text and prompt

Genau. Wo Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt. Und Charlmas erwidert Saulmas macht den Fehler, die falsche Ebene zu betrachten. Natürlich versteht der Mann im Zimmer, der nur die Regeln befolgt, kein Chinesisch. Das ist klar.

GENERAL: 
SCRIPT:
(speaker=S2, very non-narrative, very clean-bg) [0.44 seconds duration] Genau.
(speaker=S2, very non-narrative, very scattered, indifference, less-enjoyable) [0.08 seconds duration] Wo
(speaker=S1, very full, very news-anchor, concentration, enjoyable) [8.32 seconds duration] Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt.
(speaker=S2, very negative, very formal, very cynicism, enjoyable) [1.20 seconds duration] Und Charlmas erwidert
[0.52 seconds pause]
(speaker=S2, very negative, very formal, very cynicism, enjoyable) [3.60 seconds duration] Saulmas macht den Fehler, die falsche Ebene zu betrachten.
(speaker=S2, very news-anchor, very seamless-breath, enjoyable) [3.84 seconds duration] Natürlich versteht der Mann im Zimmer, der nur die Regeln befolgt, kein Chinesisch.
(speaker=S1, very accelerating, very choppy, slightly indifference) [0.56 seconds duration] Das ist klar.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.23 · genuine 3.48 · WER 7.7%
18.56 s; requested 18.56 s

CE 6.24 · genuine 3.42 · WER 46.2%
18.56 s; requested 18.56 s

CE 6.36 · genuine 3.93 · WER 1.9%
18.64 s; requested 18.56 s

CE 6.16 · genuine 3.58 · WER 13.5%
18.72 s; requested 18.56 s

de-4turn-01

Exact text and prompt

Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt. Und Charlmas erwidert Saulmas macht den Fehler, die falsche Ebene zu betrachten. Natürlich versteht der Mann im Zimmer, der nur die Regeln befolgt, kein Chinesisch. Das ist klar. Aber das gesamte System der Mensch, das Regelbuch, die Zettel, implementiert als Ganzes die erforderliche kausale Organisation. Und dieses Gesamtsystem ist es, das chinesisch versteht.

GENERAL: 
SCRIPT:
(speaker=S1, very full, very news-anchor, concentration, enjoyable) [8.20 seconds duration] Thurl ja sagt, ein System, das nur Symbole nach Regeln manipuliert, versteht die Sprache nicht wirklich, auch wenn es den Anschein erweckt.
(speaker=S2, very negative, very formal, very cynicism, enjoyable) [1.20 seconds duration] Und Charlmas erwidert
[0.52 seconds pause]
(speaker=S2, very negative, very formal, very cynicism, enjoyable) [3.60 seconds duration] Saulmas macht den Fehler, die falsche Ebene zu betrachten.
(speaker=S2, very news-anchor, very seamless-breath, enjoyable) [3.84 seconds duration] Natürlich versteht der Mann im Zimmer, der nur die Regeln befolgt, kein Chinesisch.
(speaker=S1, very accelerating, very choppy, slightly indifference) [0.56 seconds duration] Das ist klar.
(speaker=S2, very news-anchor, very oral-bright, slightly concentration, enjoyable) [8.56 seconds duration] Aber das gesamte System der Mensch, das Regelbuch, die Zettel, implementiert als Ganzes die erforderliche kausale Organisation.
(speaker=S2, very bright, very forward-mask, enjoyable) [1.52 seconds duration] Und dieses Gesamtsystem
[0.24 seconds pause]
(speaker=S2, very bright, very forward-mask, enjoyable) [2.08 seconds duration] ist es, das chinesisch versteht.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.25 · genuine 3.51 · WER 5.3%
30.40 s; requested 30.32 s

CE 6.24 · genuine 3.52 · WER 5.3%
30.24 s; requested 30.32 s

CE 6.37 · genuine 3.78 · WER 8.0%
30.32 s; requested 30.32 s

CE 6.34 · genuine 3.62 · WER 5.3%
30.16 s; requested 30.32 s

de-4turn-02

Exact text and prompt

So lautet die These. Das ist ja quasi die philosophische Lizenz zum Bau einer bewussten KI. Absolut. Eine ausreichend komplexe und korrekt organisierte Computersimulation eines menschlichen Gehirns wäre nicht nur so, als ob sie bewusst wäre, sie wäre es tatsächlich. Und das ist auch seine direkte Antwort auf John Thurrells berühmtes chinesisches Zimmerargument.

GENERAL: 
SCRIPT:
(speaker=S2, very accelerating, very tonal, slightly calmness) [1.20 seconds duration] So lautet die These.
(speaker=S1, very news-anchor, very full, slightly intrigue, enjoyable) [3.36 seconds duration] Das ist ja quasi die philosophische Lizenz zum Bau einer bewussten
[0.24 seconds pause]
(speaker=S1, very news-anchor, very full, slightly intrigue, enjoyable) [0.64 seconds duration] KI.
(speaker=S2, very smooth, very clean-bg, slightly excitement) [0.72 seconds duration] Absolut.
[0.36 seconds pause]
(speaker=S2, very news-anchor, very forward-mask, concentration, enjoyable) [5.28 seconds duration] Eine ausreichend komplexe und korrekt organisierte Computersimulation eines menschlichen Gehirns
[0.32 seconds pause]
(speaker=S2, very news-anchor, very forward-mask, concentration, enjoyable) [4.00 seconds duration] wäre nicht nur so, als ob sie bewusst wäre, sie wäre es tatsächlich.
(speaker=S1, very news-anchor, very full, attention, enjoyable) [3.12 seconds duration] Und das ist auch seine direkte Antwort auf John Thurrells berühmtes
[0.24 seconds pause]
(speaker=S1, very news-anchor, very full, attention, enjoyable) [1.72 seconds duration] chinesisches Zimmerargument.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.28 · genuine 3.61 · WER 1.9%
21.20 s; requested 21.20 s

CE 6.20 · genuine 4.01 · WER 67.9%
6.00 s; requested 21.20 s

CE 6.33 · genuine 3.84 · WER 1.9%
21.28 s; requested 21.20 s

CE 6.28 · genuine 3.89 · WER 1.9%
21.28 s; requested 21.20 s

de-4turn-03

Exact text and prompt

führt das doch unweigerlich zur Frage nach künstlicher Intelligenz? Unweigerlich? Kann eine Maschine, eine KI, bewusst sein? Und hier wird Scharmas überraschend optimistisch. Er formuliert das Prinzip der organisationalen Invarianz.

GENERAL: 
SCRIPT:
(speaker=S1, very news-anchor, very full, slightly concentration, enjoyable) [3.52 seconds duration] führt das doch unweigerlich zur Frage nach künstlicher Intelligenz?
(speaker=S2, very tonal, very clean-bg, very uncertainty) [0.80 seconds duration] Unweigerlich?
(speaker=S1, very denasal, seamless-breath, slightly curiosity, enjoyable) [2.40 seconds duration] Kann eine Maschine, eine KI, bewusst sein?
(speaker=S2, very bright, very oral-bright, very hope, enjoyable) [0.88 seconds duration] Und hier wird Scharmas
[0.24 seconds pause]
(speaker=S2, very bright, very oral-bright, very hope, enjoyable) [1.60 seconds duration] überraschend optimistisch.
(speaker=S2, very news-anchor, very bright, concentration, enjoyable) [3.36 seconds duration] Er formuliert das Prinzip der organisationalen
[0.24 seconds pause]
(speaker=S2, very news-anchor, very bright, concentration, enjoyable) [1.04 seconds duration] Invarianz.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.30 · genuine 3.70 · WER 40.0%
14.08 s; requested 14.08 s

CE 6.23 · genuine 3.58 · WER 10.0%
14.08 s; requested 14.08 s

CE 6.35 · genuine 3.70 · WER 3.3%
14.24 s; requested 14.08 s

CE 6.49 · genuine 3.96 · WER 3.3%
14.16 s; requested 14.08 s

de-4turn-04

Exact text and prompt

Jeder auf der Straße darf ihn erschlagen. Und Es ist juristisch komplett irrelevant. Gleichzeitig ist er aber nicht einmal für die Götter wertvoll genug, um als Opfertier zu dienen. Er ist völlig schutzlos. Alle Zweck

GENERAL: 
SCRIPT:
(speaker=S2, very clean-content, very seamless-breath) [2.00 seconds duration] Jeder auf der Straße darf ihn erschlagen.
(speaker=S2, bright, thin-throat, detestation, enjoyable) [0.16 seconds duration] Und
[0.64 seconds pause]
(speaker=S2, bright, thin-throat, detestation, enjoyable) [2.72 seconds duration] Es ist juristisch komplett irrelevant.
(speaker=S1, very full, very blended-resonance, very gratitude, enjoyable) [4.48 seconds duration] Gleichzeitig ist er aber nicht einmal für die Götter wertvoll genug, um als Opfertier zu dienen.
(speaker=S2, very clean-bg, very hi-fi, very bitterness) [1.36 seconds duration] Er ist völlig schutzlos.
(speaker=S1, very accelerating, very clean-bg, slightly emotional blunting) [0.56 seconds duration] Alle Zweck
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.01 · genuine 3.65 · WER 0.0%
11.92 s; requested 11.92 s

CE 5.94 · genuine 3.54 · WER 2.9%
11.92 s; requested 11.92 s

CE 6.17 · genuine 3.98 · WER 2.9%
12.00 s; requested 11.92 s

CE 6.28 · genuine 3.85 · WER 5.7%
12.08 s; requested 11.92 s

fr-2turn-00

Exact text and prompt

it isn't, what is this for? It becomes, where am I held? Where am I held?

GENERAL: 
SCRIPT:
(speaker=S2, very full, very denasal, very flabbergasted, enjoyable) [1.84 seconds duration] it isn't, what is this for?
(speaker=S2, very denasal, very full, very pensiveness, enjoyable) [2.24 seconds duration] It becomes, where am I held?
[0.24 seconds pause]
(speaker=S1, very fading, very steady, very confusion) [0.88 seconds duration] Where am I held?
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.35 · genuine 4.10 · WER 6.2%
5.12 s; requested 5.20 s

CE 6.40 · genuine 4.01 · WER 6.2%
5.20 s; requested 5.20 s

CE 6.28 · genuine 4.34 · WER 18.8%
5.36 s; requested 5.20 s

CE 6.37 · genuine 4.16 · WER 0.0%
4.96 s; requested 5.20 s

fr-2turn-01

Exact text and prompt

Ça consomme des crédits à chaque action. L'addition monte en silence, c'est pas une extension gratuite.

GENERAL: 
SCRIPT:
(speaker=S1, very clean-content, very full, slightly engrossment, enjoyable) [1.92 seconds duration] Ça consomme des crédits à chaque action.
(speaker=S2, very hi-fi, clean-bg, slightly boredom) [3.12 seconds duration] L'addition monte en silence, c'est pas une extension gratuite.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.09 · genuine 4.14 · WER 0.0%
4.96 s; requested 5.04 s

CE 6.07 · genuine 4.07 · WER 12.5%
5.04 s; requested 5.04 s

CE 6.00 · genuine 3.82 · WER 0.0%
5.20 s; requested 5.04 s

CE 5.93 · genuine 4.17 · WER 0.0%
5.04 s; requested 5.04 s

fr-2turn-02

Exact text and prompt

Batteries die. Which brings us to level three. the abide identity or the anchored identity.

GENERAL: 
SCRIPT:
(speaker=S2, very fading, very monotone, very apathy) [0.68 seconds duration] Batteries die.
(speaker=S1, very tonal, very hi-fi, slightly serenity, enjoyable) [2.08 seconds duration] Which brings us to level three.
[0.20 seconds pause]
(speaker=S1, slightly bright, slightly thin-throat, pondering, enjoyable) [3.20 seconds duration] the abide identity or the anchored identity.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.18 · genuine 4.13 · WER 0.0%
6.16 s; requested 6.16 s

CE 6.27 · genuine 4.43 · WER 0.0%
6.16 s; requested 6.16 s

CE 6.59 · genuine 3.90 · WER 6.7%
6.24 s; requested 6.16 s

CE 6.48 · genuine 4.63 · WER 20.0%
6.24 s; requested 6.16 s

fr-2turn-03

Exact text and prompt

Ils ont purement et simplement supprimé tous les titres de poste. radicale.

GENERAL: 
SCRIPT:
(speaker=S2, very news-anchor, steady, self-confidently, enjoyable) [3.40 seconds duration] Ils ont purement et simplement supprimé tous les titres de poste.
[0.44 seconds pause]
(speaker=S1, very effortless, very accelerating, numbness) [0.88 seconds duration] radicale.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.93 · genuine 4.14 · WER 16.7%
4.72 s; requested 4.72 s

CE 5.92 · genuine 4.36 · WER 8.3%
4.72 s; requested 4.72 s

CE 6.01 · genuine 3.99 · WER 8.3%
4.80 s; requested 4.72 s

CE 6.05 · genuine 4.21 · WER 16.7%
4.64 s; requested 4.72 s

fr-2turn-04

Exact text and prompt

recommandation formelle, c'est de bloquer au niveau 2. Aller plus loin demande une discipline intenable pour un individu seul.

GENERAL: 
SCRIPT:
(speaker=S1, very news-anchor, very formal, slightly absorption, enjoyable) [2.56 seconds duration] recommandation formelle, c'est de bloquer au niveau 2.
(speaker=S2, very news-anchor, very steady, slightly emotional blunting, enjoyable) [2.32 seconds duration] Aller plus loin demande une discipline intenable
[0.24 seconds pause]
(speaker=S2, very news-anchor, very steady, slightly emotional blunting, enjoyable) [1.12 seconds duration] pour un individu seul.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.90 · genuine 4.08 · WER 5.3%
6.24 s; requested 6.24 s

CE 5.93 · genuine 4.11 · WER 5.3%
6.24 s; requested 6.24 s

CE 6.40 · genuine 4.06 · WER 0.0%
6.40 s; requested 6.24 s

CE 6.39 · genuine 4.14 · WER 0.0%
6.32 s; requested 6.24 s

fr-4turn-00

Exact text and prompt

niveau 2, c'est assister, avec des agents en parallèle mais complètement isolés les uns des autres. Et on monte jusqu'où? Jusqu'au niveau 5, l'organisation native IA, où des milliers d'agents interagissent ensemble, l'autonomie totale. Sauf que la conclusion du rapport est frappante. L'éditeur pousse vers les niveaux 3 ou

GENERAL: 
SCRIPT:
(speaker=S2, very oral-bright, forward-mask, slightly concentration, enjoyable) [5.28 seconds duration] niveau 2, c'est assister, avec des agents en parallèle mais complètement isolés les uns des autres.
(speaker=S1, non-narrative, naturalistic) [0.96 seconds duration] Et on monte jusqu'où?
[0.24 seconds pause]
(speaker=S2, very news-anchor, very formal, slightly deep focus, enjoyable) [6.64 seconds duration] Jusqu'au niveau 5, l'organisation native IA, où des milliers d'agents interagissent ensemble, l'autonomie totale.
[0.24 seconds pause]
(speaker=S1, very full, very clean-bg, slightly attention, enjoyable) [2.48 seconds duration] Sauf que la conclusion du rapport est frappante.
(speaker=S1, very full, very effortless, slightly concentration, enjoyable) [2.40 seconds duration] L'éditeur pousse vers les niveaux 3 ou
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.14 · genuine 4.24 · WER 4.1%
18.16 s; requested 18.24 s

CE 6.01 · genuine 4.18 · WER 10.2%
18.16 s; requested 18.24 s

CE 6.21 · genuine 3.94 · WER 2.0%
18.16 s; requested 18.24 s

CE 6.15 · genuine 4.18 · WER 4.1%
18.16 s; requested 18.24 s

fr-4turn-01

Exact text and prompt

en pure coordination. 50% ? Oui. La moitié du temps juste pour se coordonner.

GENERAL: 
SCRIPT:
(speaker=S2, very clean-bg, smooth) [1.12 seconds duration] en pure coordination.
[0.24 seconds pause]
(speaker=S1, accelerating, clean-bg) [0.96 seconds duration] 50% ?
(speaker=S2, very non-narrative, very decelerating, slightly emotional blunting, less-enjoyable) [0.24 seconds duration] Oui.
(speaker=S1, very clean-bg, hi-fi, enjoyable) [2.64 seconds duration] La moitié du temps juste pour se coordonner.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.06 · genuine 4.28 · WER 0.0%
5.20 s; requested 5.20 s

CE 6.07 · genuine 4.29 · WER 23.1%
5.20 s; requested 5.20 s

CE 6.26 · genuine 4.34 · WER 23.1%
5.44 s; requested 5.20 s

CE 6.05 · genuine 4.03 · WER 0.0%
5.36 s; requested 5.20 s

fr-4turn-02

Exact text and prompt

So the branch is just the conduit. It's just a conduit. The life is received, not self-generated. The notes link this to the idea of being sent, and they use this phrase, the orphan identity. This is the opposite of abiding.

GENERAL: 
SCRIPT:
(speaker=S1, clean-bg, thin-throat, enjoyable) [2.36 seconds duration] So the branch is just the conduit.
(speaker=S2, very accelerating, very clean-bg, very apathy) [1.20 seconds duration] It's just a conduit.
(speaker=S2, very full, very monologic, very insensitivity, enjoyable) [2.80 seconds duration] The life is received, not self-generated.
(speaker=S1, very clean-bg, very tonal, deep focus, enjoyable) [1.52 seconds duration] The notes link this
[0.24 seconds pause]
(speaker=S1, very clean-bg, very tonal, deep focus, enjoyable) [3.84 seconds duration] to the idea of being sent, and they use this phrase, the orphan
[0.24 seconds pause]
(speaker=S1, very clean-bg, very tonal, deep focus, enjoyable) [0.64 seconds duration] identity.
(speaker=S2, very hi-fi, very clean-content, enjoyable) [1.72 seconds duration] This is the opposite of abiding.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.18 · genuine 3.80 · WER 0.0%
14.56 s; requested 14.56 s

CE 6.17 · genuine 3.56 · WER 0.0%
14.56 s; requested 14.56 s

CE 6.26 · genuine 3.94 · WER 23.8%
15.12 s; requested 14.56 s

CE 6.50 · genuine 4.19 · WER 9.5%
14.96 s; requested 14.56 s

fr-4turn-03

Exact text and prompt

50% ? Oui. La moitié du temps juste pour se coordonner. Ouais, la moitié, c'est la taxe de synchronisation.

GENERAL: 
SCRIPT:
(speaker=S1, accelerating, clean-bg) [0.96 seconds duration] 50% ?
(speaker=S2, very non-narrative, very decelerating, slightly emotional blunting, less-enjoyable) [0.24 seconds duration] Oui.
(speaker=S1, very clean-bg, hi-fi, enjoyable) [2.64 seconds duration] La moitié du temps juste pour se coordonner.
(speaker=S2, clean-bg, hi-fi, slightly relaxation, enjoyable) [3.04 seconds duration] Ouais, la moitié, c'est la taxe de synchronisation.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.28 · genuine 4.14 · WER 0.0%
6.80 s; requested 6.88 s

CE 6.36 · genuine 4.31 · WER 50.0%
6.80 s; requested 6.88 s

CE 6.24 · genuine 4.11 · WER 0.0%
7.04 s; requested 6.88 s

CE 6.10 · genuine 4.06 · WER 0.0%
6.96 s; requested 6.88 s

fr-4turn-04

Exact text and prompt

I can say I'm trusting, but my stomach might still be in knots. Right. The deep dive points to James chapter three as the diagnostic. James 3.2.

GENERAL: 
SCRIPT:
(speaker=S1, very darkening, smooth, doubt, enjoyable) [2.92 seconds duration] I can say I'm trusting, but my stomach might still be in knots.
(speaker=S2, very fading, very scattered, slightly emotional blunting, less-enjoyable) [0.40 seconds duration] Right.
(speaker=S1, very news-anchor, very seamless-breath, slightly engrossment, enjoyable) [3.44 seconds duration] The deep dive points to James chapter three as the diagnostic.
(speaker=S2, very naturalistic, accelerating, slightly lightness) [1.32 seconds duration] James 3.2.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.25 · genuine 4.12 · WER 0.0%
8.08 s; requested 8.08 s

CE 6.15 · genuine 4.28 · WER 60.7%
8.08 s; requested 8.08 s

CE 6.28 · genuine 3.93 · WER 32.1%
8.16 s; requested 8.08 s

CE 6.22 · genuine 3.85 · WER 10.7%
8.24 s; requested 8.08 s

es-2turn-00

Exact text and prompt

Exacto. Pero mientras tanto, en nuestro juego de números, un cuerpo vale poco más de un millón de euros.

GENERAL: 
SCRIPT:
(speaker=S2, very accelerating, very effortless, slightly stoicism) [0.56 seconds duration] Exacto.
(speaker=S1, long-phrased, even-rhythm, slightly intrigue, enjoyable) [5.12 seconds duration] Pero mientras tanto, en nuestro juego de números, un cuerpo vale poco más de un millón de euros.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.87 · genuine 3.91 · WER 0.0%
5.68 s; requested 5.68 s

CE 5.90 · genuine 4.01 · WER 0.0%
5.68 s; requested 5.68 s

CE 6.10 · genuine 4.14 · WER 0.0%
5.68 s; requested 5.68 s

CE 6.05 · genuine 3.96 · WER 0.0%
5.76 s; requested 5.68 s

es-2turn-01

Exact text and prompt

a las que apodaron los astronautas subterráneos. Exacto.

GENERAL: 
SCRIPT:
(speaker=S1, very hi-fi, very bright, very self-consciousness, enjoyable) [0.88 seconds duration] a las que apodaron
[0.24 seconds pause]
(speaker=S1, very hi-fi, very bright, very self-consciousness, enjoyable) [1.92 seconds duration] los astronautas subterráneos.
(speaker=S2, very accelerating, non-narrative, slightly emotional blunting) [0.56 seconds duration] Exacto.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.37 · genuine 4.36 · WER 0.0%
3.60 s; requested 3.60 s

CE 5.62 · genuine 4.34 · WER 12.5%
3.60 s; requested 3.60 s

CE 6.21 · genuine 4.44 · WER 0.0%
4.24 s; requested 3.60 s

CE 6.13 · genuine 4.49 · WER 25.0%
3.36 s; requested 3.60 s

es-2turn-02

Exact text and prompt

Prácticamente. Eso significa que estaban por ahí al mismo tiempo que nuestros primeros ancestros directos, nuestra propia especie.

GENERAL: 
SCRIPT:
(speaker=S2, very accelerating, denasal, detachment) [0.88 seconds duration] Prácticamente.
(speaker=S1, very clean-bg, bright, very dignity, enjoyable) [6.16 seconds duration] Eso significa que estaban por ahí al mismo tiempo que nuestros primeros ancestros directos, nuestra propia especie.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.13 · genuine 4.13 · WER 0.0%
7.04 s; requested 7.04 s

CE 6.23 · genuine 3.96 · WER 0.0%
7.04 s; requested 7.04 s

CE 6.27 · genuine 3.92 · WER 0.0%
7.04 s; requested 7.04 s

CE 5.82 · genuine 3.87 · WER 0.0%
6.56 s; requested 7.04 s

es-2turn-03

Exact text and prompt

335 mil. Eso es.

GENERAL: 
SCRIPT:
(speaker=S1, very hi-fi, very clean-bg) [1.60 seconds duration] 335 mil.
(speaker=S2, very accelerating, very non-narrative) [0.72 seconds duration] Eso es.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.38 · genuine 4.26 · WER 100.0%
2.32 s; requested 2.32 s

CE 5.38 · genuine 4.24 · WER 25.0%
2.32 s; requested 2.32 s

CE 5.83 · genuine 4.32 · WER 100.0%
4.16 s; requested 2.32 s

CE 5.44 · genuine 4.41 · WER 100.0%
2.40 s; requested 2.32 s

es-2turn-04

Exact text and prompt

restos de Homonaledi tienen entre 335.000 y 236.000 años. Un momento, para

GENERAL: 
SCRIPT:
(speaker=S2, very full, very news-anchor, slightly engrossment, enjoyable) [5.44 seconds duration] restos de Homonaledi tienen entre 335.000 y 236.000 años.
(speaker=S1, very clean-bg, very tonal, slightly fervor) [0.80 seconds duration] Un momento, para
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.34 · genuine 4.21 · WER 21.4%
6.16 s; requested 6.24 s

CE 6.24 · genuine 4.06 · WER 42.9%
6.24 s; requested 6.24 s

CE 6.22 · genuine 3.88 · WER 7.1%
6.24 s; requested 6.24 s

CE 6.02 · genuine 4.17 · WER 100.0%
6.08 s; requested 6.24 s

es-4turn-00

Exact text and prompt

335 mil. Eso es. Ayer, en términos evolutivos. Prácticamente.

GENERAL: 
SCRIPT:
(speaker=S1, very hi-fi, very clean-bg) [1.60 seconds duration] 335 mil.
(speaker=S2, very accelerating, very non-narrative) [0.72 seconds duration] Eso es.
(speaker=S1, very clean-content, very bright, slightly dignity, enjoyable) [1.84 seconds duration] Ayer, en términos evolutivos.
(speaker=S2, very accelerating, denasal, detachment) [1.04 seconds duration] Prácticamente.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.90 · genuine 4.08 · WER 11.1%
5.20 s; requested 5.20 s

CE 5.89 · genuine 4.04 · WER 11.1%
5.20 s; requested 5.20 s

CE 6.01 · genuine 4.02 · WER 11.1%
5.44 s; requested 5.20 s

CE 6.02 · genuine 3.96 · WER 11.1%
5.28 s; requested 5.20 s

es-4turn-01

Exact text and prompt

Que el capitalismo no valora a la persona, valora la función o el recurso. En este caso, la capacidad biológica de crear vida y un estándar de belleza muy concreto tienen un precio más alto. Que cualquier otra diferencia, al final, lo que hemos hecho es encontrar diferentes tipos de

GENERAL: 
SCRIPT:
(speaker=S2, very full, very news-anchor, absorption, enjoyable) [4.40 seconds duration] Que el capitalismo no valora a la persona, valora la función o el recurso.
(speaker=S1, very news-anchor, very clean-bg, ease, enjoyable) [6.32 seconds duration] En este caso, la capacidad biológica de crear vida y un estándar de belleza muy concreto tienen un precio más alto.
(speaker=S2, clean-content, denasal) [1.44 seconds duration] Que cualquier otra diferencia,
[0.40 seconds pause]
(speaker=S1, very news-anchor, very blended-resonance, slightly intrigue) [3.04 seconds duration] al final, lo que hemos hecho es encontrar diferentes tipos de
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.12 · genuine 3.93 · WER 2.0%
15.68 s; requested 15.60 s

CE 6.13 · genuine 3.82 · WER 2.0%
15.76 s; requested 15.60 s

CE 6.20 · genuine 4.13 · WER 0.0%
15.68 s; requested 15.60 s

CE 6.09 · genuine 3.61 · WER 98.0%
15.68 s; requested 15.60 s

es-4turn-02

Exact text and prompt

Vale, empecemos por la contradicción central, el cráneo. Pues tenía un cerebro diminuto, entre 465 y 560 centímetros cúbicos. Que eso, para que nos hagamos una idea, es. apenas un tercio del nuestro.

GENERAL: 
SCRIPT:
(speaker=S1, very bright, guarded, slightly concentration, enjoyable) [3.36 seconds duration] Vale, empecemos por la contradicción central, el cráneo.
(speaker=S2, very full, very news-anchor, indifference, enjoyable) [1.12 seconds duration] Pues tenía un cerebro
[0.24 seconds pause]
(speaker=S2, very full, very news-anchor, indifference, enjoyable) [2.56 seconds duration] diminuto, entre 465
[0.24 seconds pause]
(speaker=S2, very full, very news-anchor, indifference, enjoyable) [2.24 seconds duration] y 560 centímetros cúbicos.
(speaker=S1, slightly dialogic, slightly forward-mask, skepticism, enjoyable) [2.48 seconds duration] Que eso, para que nos hagamos una idea, es.
[0.32 seconds pause]
(speaker=S2, very full, very steady, very resentment, enjoyable) [0.08 seconds duration] apenas
[0.32 seconds pause]
(speaker=S2, very full, very steady, very resentment, enjoyable) [1.60 seconds duration] un tercio del nuestro.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.10 · genuine 3.83 · WER 0.0%
14.56 s; requested 14.56 s

CE 6.06 · genuine 4.03 · WER 0.0%
14.56 s; requested 14.56 s

CE 6.40 · genuine 4.39 · WER 3.0%
14.72 s; requested 14.56 s

CE 6.29 · genuine 4.15 · WER 0.0%
14.56 s; requested 14.56 s

es-4turn-03

Exact text and prompt

Eso es. Ayer, en términos evolutivos. Prácticamente. Eso significa que estaban por ahí al mismo tiempo que nuestros primeros ancestros directos, nuestra propia especie.

GENERAL: 
SCRIPT:
(speaker=S2, very accelerating, very non-narrative) [0.64 seconds duration] Eso es.
(speaker=S1, very clean-content, very bright, slightly dignity, enjoyable) [1.84 seconds duration] Ayer, en términos evolutivos.
(speaker=S2, very accelerating, denasal, detachment) [1.04 seconds duration] Prácticamente.
(speaker=S1, very clean-bg, bright, very dignity, enjoyable) [6.16 seconds duration] Eso significa que estaban por ahí al mismo tiempo que nuestros primeros ancestros directos, nuestra propia especie.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.00 · genuine 3.79 · WER 0.0%
9.60 s; requested 9.68 s

CE 5.94 · genuine 4.00 · WER 0.0%
9.68 s; requested 9.68 s

CE 6.17 · genuine 4.12 · WER 0.0%
9.68 s; requested 9.68 s

CE 6.09 · genuine 4.17 · WER 0.0%
9.68 s; requested 9.68 s

es-4turn-04

Exact text and prompt

esas zonas para qué sirven? Son zonas que en los humanos modernos asociamos con la planificación, la toma de decisiones, incluso potencialmente con el lenguaje. Un cerebro pequeño, pero con una organización interna inesperadamente compleja. Justo.

GENERAL: 
SCRIPT:
(speaker=S1, very accelerating, very clean-bg, slightly inspiration, enjoyable) [1.40 seconds duration] esas zonas para qué sirven?
(speaker=S2, very full, very news-anchor, absorption, enjoyable) [7.28 seconds duration] Son zonas que en los humanos modernos asociamos con la planificación, la toma de decisiones, incluso potencialmente con el lenguaje.
(speaker=S1, bright, hi-fi, slightly intrigue, enjoyable) [4.64 seconds duration] Un cerebro pequeño, pero con una organización interna inesperadamente compleja.
(speaker=S2, very flat, steady) [0.60 seconds duration] Justo.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.17 · genuine 3.87 · WER 0.0%
13.92 s; requested 13.92 s

CE 6.16 · genuine 3.96 · WER 0.0%
13.92 s; requested 13.92 s

CE 6.38 · genuine 4.09 · WER 0.0%
13.92 s; requested 13.92 s

CE 6.26 · genuine 3.82 · WER 2.8%
13.92 s; requested 13.92 s

nl-2turn-00

Exact text and prompt

Kijk, als een soort in de rest van Nederland al bijna is uitgestorven, maar in het noorden nog net overleeft, dan moet je wel.

GENERAL: 
SCRIPT:
(speaker=S2, very bright, very oral-bright, fascination, enjoyable) [7.04 seconds duration] Kijk, als een soort in de rest van Nederland al bijna is uitgestorven, maar in het noorden nog net overleeft, dan moet je wel.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.13 · genuine 4.04 · WER 0.0%
7.04 s; requested 7.04 s

CE 5.91 · genuine 4.07 · WER 0.0%
7.04 s; requested 7.04 s

CE 6.25 · genuine 3.94 · WER 0.0%
7.04 s; requested 7.04 s

CE 5.62 · genuine 4.57 · WER 0.0%
6.96 s; requested 7.04 s

nl-2turn-01

Exact text and prompt

Zeker. En Tom en Lisa slikken hun trots in. Ze accepteren de grens die de natuur trekt en kiezen voor veiligheid. Pragmatisch kopen ze maar een houten olifantje en een rode sjaal. En er was nog een klein verhaal over een havenwerker,

GENERAL: 
SCRIPT:
(speaker=S2, very non-narrative, very flat, very uncertainty) [0.64 seconds duration] Zeker.
(speaker=S2, very full, very narrator, slightly deep focus, enjoyable) [1.12 seconds duration] En Tom en Lisa
[0.48 seconds pause]
(speaker=S2, very full, very narrator, slightly deep focus, enjoyable) [1.36 seconds duration] slikken hun trots in.
(speaker=S2, very news-anchor, very full, slightly deep focus, enjoyable) [2.56 seconds duration] Ze accepteren de grens die de natuur trekt
[0.24 seconds pause]
(speaker=S2, very news-anchor, very full, slightly deep focus, enjoyable) [1.76 seconds duration] en kiezen voor veiligheid.
(speaker=S2, very full, very seamless-breath, enjoyable) [4.00 seconds duration] Pragmatisch kopen ze maar een houten olifantje en een rode sjaal.
(speaker=S1, very forward-mask, very bright, slightly interest, enjoyable) [3.12 seconds duration] En er was nog een klein verhaal over een havenwerker,
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.19 · genuine 3.92 · WER 0.0%
15.36 s; requested 15.28 s

CE 6.12 · genuine 4.12 · WER 2.4%
15.36 s; requested 15.28 s

CE 6.23 · genuine 3.79 · WER 2.4%
15.36 s; requested 15.28 s

CE 6.25 · genuine 3.92 · WER 2.4%
15.36 s; requested 15.28 s

nl-2turn-02

Exact text and prompt

Ja, de fysieke omgeving is daar steriel, leeg, er is totale controle, er is letterlijk geen wrijving. Exact. En aan de andere kant.

GENERAL: 
SCRIPT:
(speaker=S1, very oral-bright, very bright, very fascination, enjoyable) [7.36 seconds duration] Ja, de fysieke omgeving is daar steriel, leeg, er is totale controle, er is letterlijk geen wrijving.
(speaker=S2, very non-narrative, very accelerating, detachment) [0.48 seconds duration] Exact.
(speaker=S2, very accelerating, very clean-content, enjoyable) [1.20 seconds duration] En aan de andere kant.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.59 · genuine 4.19 · WER 4.3%
9.04 s; requested 9.04 s

CE 5.69 · genuine 4.20 · WER 8.7%
9.04 s; requested 9.04 s

CE 6.11 · genuine 4.11 · WER 0.0%
9.12 s; requested 9.04 s

CE 6.06 · genuine 4.31 · WER 4.3%
9.12 s; requested 9.04 s

nl-2turn-03

Exact text and prompt

Zeker, en dat laat ons eigenlijk achter met één hele intrigerende laatste gedachte om even op te kouwen. Ik ben benieuwd.

GENERAL: 
SCRIPT:
(speaker=S2, very clean-content, very forward-mask, intrigue, enjoyable) [2.64 seconds duration] Zeker, en dat laat ons eigenlijk achter met één hele
[0.24 seconds pause]
(speaker=S2, very clean-content, very forward-mask, intrigue, enjoyable) [3.12 seconds duration] intrigerende laatste gedachte om even op te kouwen.
(speaker=S1, flat-resonance, monologic) [0.72 seconds duration] Ik ben benieuwd.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 5.59 · genuine 3.91 · WER 0.0%
6.72 s; requested 6.72 s

CE 5.59 · genuine 3.84 · WER 0.0%
6.72 s; requested 6.72 s

CE 6.10 · genuine 4.18 · WER 0.0%
6.72 s; requested 6.72 s

CE 6.14 · genuine 3.94 · WER 0.0%
6.64 s; requested 6.72 s

nl-2turn-04

Exact text and prompt

we willen onszelf gewoon heel graag wijsmaken dat het goud in de god ligt of dat de auto het allemaal wel oplost zodat we heel even onze ogen dicht kunnen doen. Dat is een hele scherpe observatie en de psychologie ondersteunt. Steunt die behoefte aan rust ook volledig. We noemen dat cognitieve uitputting.

GENERAL: 
SCRIPT:
(speaker=S1, very news-anchor, very bright, fervor, enjoyable) [2.40 seconds duration] we willen onszelf gewoon heel graag wijsmaken
[0.24 seconds pause]
(speaker=S1, very news-anchor, very bright, fervor, enjoyable) [1.60 seconds duration] dat het goud in de god ligt
[0.32 seconds pause]
(speaker=S1, very news-anchor, very bright, fervor, enjoyable) [5.20 seconds duration] of dat de auto het allemaal wel oplost zodat we heel even onze ogen dicht kunnen doen.
(speaker=S2, very full, very news-anchor, slightly engrossment, enjoyable) [3.60 seconds duration] Dat is een hele scherpe observatie en de psychologie ondersteunt.
[0.24 seconds pause]
(speaker=S2, very news-anchor, very formal, enjoyable) [2.32 seconds duration] Steunt die behoefte aan rust ook volledig.
(speaker=S2, very full, very clean-content, slightly engrossment, enjoyable) [2.64 seconds duration] We noemen dat cognitieve uitputting.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.08 · genuine 3.62 · WER 0.0%
18.56 s; requested 18.56 s

CE 6.02 · genuine 3.73 · WER 3.8%
18.56 s; requested 18.56 s

CE 6.05 · genuine 3.61 · WER 0.0%
18.64 s; requested 18.56 s

CE 5.89 · genuine 3.48 · WER 5.7%
18.64 s; requested 18.56 s

nl-4turn-00

Exact text and prompt

zodat ze zelf thee kan maken en minder plastic flessen hoeft te kopen. Ja, een hele tastbare actie. En Milan die pakt dikke stiften en schrijft een bordje voor in de etalage. Neem een kleine groene stap vandaag. Even later verkoopt Moer nog een plant die subtiel naar citroen ruikt aan een jongen, specifiek om zijn overwerkte moeder wat rust te geven. En daar zie je dus die verschuiving. 20 graden voelt onoverkomelijk, maar de focus verleggen naar de vensterbank van het 12jarige, dat herstelt direct een gevoel van controle en daadkracht.

GENERAL: 
SCRIPT:
(speaker=S1, very forward-mask, very oral-bright, slightly fascination, enjoyable) [1.84 seconds duration] zodat ze zelf thee kan maken
[0.24 seconds pause]
(speaker=S1, very forward-mask, very oral-bright, slightly fascination, enjoyable) [2.40 seconds duration] en minder plastic flessen hoeft te kopen.
(speaker=S2, very dialogic, very emphatic, interest, enjoyable) [2.20 seconds duration] Ja, een hele tastbare actie.
(speaker=S1, very hi-fi, very oral-bright, slightly fulfillment, enjoyable) [5.28 seconds duration] En Milan die pakt dikke stiften en schrijft een bordje voor in de etalage.
(speaker=S1, very bright, very hi-fi, slightly joviality, enjoyable) [3.12 seconds duration] Neem een kleine groene stap vandaag.
(speaker=S1, very news-anchor, very bright, slightly deep focus, enjoyable) [4.72 seconds duration] Even later verkoopt Moer nog een plant die subtiel naar citroen ruikt
[0.32 seconds pause]
(speaker=S1, very news-anchor, very bright, slightly deep focus, enjoyable) [4.32 seconds duration] aan een jongen, specifiek om zijn overwerkte moeder wat rust te geven.
(speaker=S2, very full, very clean-content, enjoyable) [2.68 seconds duration] En daar zie je dus die verschuiving.
(speaker=S2, very news-anchor, very full, engrossment, enjoyable) [8.48 seconds duration] 20 graden voelt onoverkomelijk, maar de focus verleggen naar de vensterbank van het 12jarige, dat herstelt direct een gevoel van controle en daadkracht.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.04 · genuine 3.50 · WER 2.2%
35.44 s; requested 35.60 s

CE 6.09 · genuine 3.52 · WER 5.4%
35.36 s; requested 35.60 s

CE 6.16 · genuine 3.68 · WER 2.2%
35.60 s; requested 35.60 s

CE 6.22 · genuine 3.44 · WER 3.3%
35.60 s; requested 35.60 s

nl-4turn-01

Exact text and prompt

Ja, een hele tastbare actie. En Milan die pakt dikke stiften en schrijft een bordje voor in de etalage. Neem een kleine groene stap vandaag. Even later verkoopt Moer nog een plant die subtiel naar citroen ruikt aan een jongen, specifiek om zijn overwerkte moeder wat rust te geven. En daar zie je dus die verschuiving. 20 graden voelt onoverkomelijk, maar de focus verleggen naar de vensterbank van het 12jarige, dat herstelt direct een gevoel van controle en daadkracht. En we zien exact hetzelfde mechanisme van die radicale praktische schaalverkleining terug in een heel ander domein in de teksten, namelijk de oudere zorg in Nijmegen.

GENERAL: 
SCRIPT:
(speaker=S2, very dialogic, very emphatic, interest, enjoyable) [2.20 seconds duration] Ja, een hele tastbare actie.
(speaker=S1, very hi-fi, very oral-bright, slightly fulfillment, enjoyable) [5.28 seconds duration] En Milan die pakt dikke stiften en schrijft een bordje voor in de etalage.
(speaker=S1, very bright, very hi-fi, slightly joviality, enjoyable) [3.12 seconds duration] Neem een kleine groene stap vandaag.
(speaker=S1, very news-anchor, very bright, slightly deep focus, enjoyable) [4.72 seconds duration] Even later verkoopt Moer nog een plant die subtiel naar citroen ruikt
[0.32 seconds pause]
(speaker=S1, very news-anchor, very bright, slightly deep focus, enjoyable) [4.32 seconds duration] aan een jongen, specifiek om zijn overwerkte moeder wat rust te geven.
(speaker=S2, very full, very clean-content, enjoyable) [2.68 seconds duration] En daar zie je dus die verschuiving.
(speaker=S2, very news-anchor, very full, engrossment, enjoyable) [8.48 seconds duration] 20 graden voelt onoverkomelijk, maar de focus verleggen naar de vensterbank van het 12jarige, dat herstelt direct een gevoel van controle en daadkracht.
(speaker=S1, very news-anchor, very bright, absorption, enjoyable) [5.28 seconds duration] En we zien exact hetzelfde mechanisme van die radicale praktische schaalverkleining
[0.24 seconds pause]
(speaker=S1, very news-anchor, very bright, absorption, enjoyable) [5.20 seconds duration] terug in een heel ander domein in de teksten, namelijk de oudere zorg in Nijmegen.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.02 · genuine 3.42 · WER 1.9%
41.76 s; requested 41.84 s

CE 6.10 · genuine 3.65 · WER 4.8%
41.92 s; requested 41.84 s

CE 6.13 · genuine 3.66 · WER 3.8%
41.76 s; requested 41.84 s

CE 6.02 · genuine 3.52 · WER 8.6%
41.28 s; requested 41.84 s

nl-4turn-02

Exact text and prompt

Ik ben benieuwd. Nou, we hebben in deze reconstructie gezien dat het absolute fundament voor de bloei en de wederopstanding van Groningen lag in het omarmen van wat we vroeger rommel noemden. Ja, de dode takken, de brandnetels. Precies, de schijnbare chaos. Het bleek uiteindelijk de kraamkamer van al het nieuwe leven te

GENERAL: 
SCRIPT:
(speaker=S1, flat-resonance, monologic) [0.72 seconds duration] Ik ben benieuwd.
(speaker=S2, very forward-mask, very tonal, hope, enjoyable) [2.56 seconds duration] Nou, we hebben in deze reconstructie gezien
[0.24 seconds pause]
(speaker=S2, very forward-mask, very tonal, hope, enjoyable) [0.96 seconds duration] dat het absolute
[0.24 seconds pause]
(speaker=S2, very forward-mask, very tonal, hope, enjoyable) [5.52 seconds duration] fundament voor de bloei en de wederopstanding van Groningen lag in het omarmen van wat we vroeger
[0.24 seconds pause]
(speaker=S2, very forward-mask, very tonal, hope, enjoyable) [1.12 seconds duration] rommel noemden.
(speaker=S1, very full, very clean-bg, slightly bantering, enjoyable) [2.16 seconds duration] Ja, de dode takken, de brandnetels.
(speaker=S2, very sharp-onset, very metallic, slightly wrath) [2.40 seconds duration] Precies, de schijnbare chaos.
(speaker=S2, very bright, very hi-fi, slightly interest, enjoyable) [3.68 seconds duration] Het bleek uiteindelijk de kraamkamer van al het nieuwe leven te
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.12 · genuine 3.97 · WER 3.8%
19.84 s; requested 19.84 s

CE 6.08 · genuine 3.85 · WER 13.2%
19.92 s; requested 19.84 s

CE 6.10 · genuine 3.92 · WER 7.5%
20.00 s; requested 19.84 s

CE 6.14 · genuine 4.33 · WER 5.7%
19.92 s; requested 19.84 s

nl-4turn-03

Exact text and prompt

en zo stort het hele kaartenhuis in elkaar. Ah, en om dat te doorbreken introduceerde de gemeente Groningen dat concept van de 50 van 050. Ja, en dat was een direct vervolg op een vergelijkbaar heel succesvol project in Eindhoven destijds. Het doel was dus een heel streng geselecteerde lijst van 50 inheemse ecologisch cruciale plantensoorten

GENERAL: 
SCRIPT:
(speaker=S2, very hi-fi, very narrator, slightly interest, enjoyable) [2.64 seconds duration] en zo stort het hele kaartenhuis in elkaar.
[0.28 seconds pause]
(speaker=S1, very full, very news-anchor, attention, enjoyable) [6.80 seconds duration] Ah, en om dat te doorbreken introduceerde de gemeente Groningen dat concept van de 50 van 050.
(speaker=S2, very bright, very oral-bright, very hope, enjoyable) [6.24 seconds duration] Ja, en dat was een direct vervolg op een vergelijkbaar heel succesvol project in Eindhoven destijds.
(speaker=S1, very full, very news-anchor, slightly intrigue, enjoyable) [7.24 seconds duration] Het doel was dus een heel streng geselecteerde lijst van 50 inheemse ecologisch cruciale plantensoorten
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.14 · genuine 3.63 · WER 3.6%
23.12 s; requested 23.20 s

CE 6.05 · genuine 3.61 · WER 7.1%
23.20 s; requested 23.20 s

CE 6.12 · genuine 3.47 · WER 0.0%
23.20 s; requested 23.20 s

CE 6.01 · genuine 3.42 · WER 1.8%
23.28 s; requested 23.20 s

nl-4turn-04

Exact text and prompt

de uitleg van Asja snijdt echt dialect hout. Wat zegt ze precies? Asja legt uit, als we deze mannen, die hun hele leven zelf een keuze hebben gemaakt, nu ineens alles blind gaan verbieden vanuit de regeltjes van het instituut, dan kruipen ze stiekem door de poort naar de straat. In een zorginstelling wordt geplaatst, treedt vaak institutionalisering op.

GENERAL: 
SCRIPT:
(speaker=S1, very hi-fi, very tonal, slightly curiosity, enjoyable) [1.04 seconds duration] de uitleg van Asja
[0.32 seconds pause]
(speaker=S1, very hi-fi, very tonal, slightly curiosity, enjoyable) [1.60 seconds duration] snijdt echt dialect hout.
(speaker=S2, very non-narrative, tonal, very doubt) [0.96 seconds duration] Wat zegt ze precies?
(speaker=S1, very bright, very oral-bright, very disapproval, enjoyable) [11.36 seconds duration] Asja legt uit, als we deze mannen, die hun hele leven zelf een keuze hebben gemaakt, nu ineens alles blind gaan verbieden vanuit de regeltjes van het instituut, dan kruipen ze stiekem door de poort naar de straat.
(speaker=S2, very news-anchor, very full, slightly engrossment, enjoyable) [4.56 seconds duration] In een zorginstelling wordt geplaatst, treedt vaak institutionalisering op.
S1 identity reference
S2 identity reference
Earlier conversation
Held-out source continuation
NotebookLM full FT · baseNotebookLM + rank 64 · final 652Acting High 8,654 · baseActing High 8,654 + rank 64 · final 652

CE 6.00 · genuine 3.50 · WER 0.0%
19.84 s; requested 19.84 s

CE 5.97 · genuine 3.49 · WER 6.8%
19.92 s; requested 19.84 s

CE 6.23 · genuine 3.61 · WER 1.7%
19.92 s; requested 19.84 s

CE 6.10 · genuine 3.63 · WER 1.7%
19.84 s; requested 19.84 s