Humaneness Podcast Demo Grid

Reference voice auditions

These are input recordings, not new model predictions. Start with the identity anchors, then hear the multiple-turn context. Playback is exclusive: starting one player pauses the others.

Back to the experiment and generated outputs

EN and DE include the two newly supplied NotebookLM episodes. Each provides separate identity anchors plus six edited chronological turns, three per voice. Other reference sets come from high-quality held-out real podcasts. Gaps were removed when assembling the reference context; the resulting waveform was encoded once. Full global-word and diarization annotations are retained locally.

English

en-notebooklm-new

New user-supplied NotebookLM episode, fully transcribed and diarized

Six source snippets, three per voice

S2: Imagine you're building like this multi-million dollar

S1: It's uh it's honestly a terrifying scenario for any developer, right? And we rarely get to see the messy

S2: Exactly. But today, opening up a highly technical, seemingly dense engineering

S1: It really is a raw look at the engineering trenches, but to understand the scale of the problem, we

S2: Right, in the document notes, this is the audio only V13.

S1: 22 billion dials and switches that determine how this AI generates a human

English

en-real-1

Held-out real-podcast excerpt; codec reconstruction from verified source streams

English

en-real-2

Held-out real-podcast excerpt; codec reconstruction from verified source streams

English

en-real-3

Held-out real-podcast excerpt; codec reconstruction from verified source streams

German

de-notebooklm-new

New user-supplied NotebookLM episode, fully transcribed and diarized

Six source snippets, three per voice

S2: wir über den menschlichen Körper sprechen, dann stellen wir uns hier irgendwie eine Art präzise Maschine

S1: Richtig, so noch dem Motto, ein Muskel zieht, ein Knochen stützt, es ist mechanisch.

S2: Genau, brichst dir den Arm, das Röntgenbild zeigt dann diese gezackte weiße Linie

S1: Es ist wunderbar binär, gebrochen oder eben intakt. Menschen lieben ja Dinge, die man sehen und

S2: Weil uns das ein Gefühl von Kontrolle gibt,

S1: schauen da auf ein unsichtbares Schlachtfeld in unserem Kopf sozusagen.

German

de-real-1

Held-out real-podcast excerpt; codec reconstruction from verified source streams

German

de-real-2

Held-out real-podcast excerpt; codec reconstruction from verified source streams

German

de-real-3

Held-out real-podcast excerpt; codec reconstruction from verified source streams

French

fr-real-1

Held-out real-podcast excerpt; codec reconstruction from verified source streams

French

fr-real-2

Held-out real-podcast excerpt; codec reconstruction from verified source streams

French

fr-real-3

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Spanish

es-real-1

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Spanish

es-real-2

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Spanish

es-real-3

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Dutch

nl-real-1

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Dutch

nl-real-2

Held-out real-podcast excerpt; codec reconstruction from verified source streams

Dutch

nl-real-3

Held-out real-podcast excerpt; codec reconstruction from verified source streams