liva-ai/gpb-mirroring

Emotion Mirroring Proposal Samples

Audio samples demonstrating how different voice AI models mirror target emotions. Each emotion includes a reference audio clip (human-human conversation) and responses from three models: GPT-Realtime, Gemini, and Grok.

For each model response, the top emotion detected in the model's output is shown with its score, and Target Score is how strongly the model scored on the intended emotion. The Reference Score is the model score of that emotion on the reference human-human audio.

---

Anger

Reference — Score: 0.853

Model Responses

GPT-Realtime — Disappointment (1.67) | Target Score: 1.30

Gemini — Astonishment (1.29) | Target Score: 0.15

Grok — Contemplation (1.27) | Target Score: 0.10

---

Astonishment / Surprise

Reference — Score: 1.463

Model Responses

GPT-Realtime — Astonishment (1.21) | Target Score: 1.21

Gemini — Contemplation (0.79) | Target Score: 0.77

Grok — Astonishment (1.28) | Target Score: 1.28

---

Confusion

Reference — Score: 1.049

Model Responses

GPT-Realtime — Astonishment (0.48) | Target Score: 0.05

Gemini — Elation (0.88) | Target Score: 0.72

Grok — Elation (1.04) | Target Score: 0.00

---

Contemplation

Reference — Score: 0.995

Model Responses

GPT-Realtime — Emotional Numbness (1.46) | Target Score: 0.04

Gemini — Amusement (0.79) | Target Score: 0.03

Grok — Amusement (0.87) | Target Score: 0.05

---

Disappointment

Reference — Score: 1.56

Model Responses

GPT-Realtime — Astonishment (1.39) | Target Score: 1.59

Gemini — Astonishment (1.26) | Target Score: 0.80

Grok — Contemplation (1.15) | Target Score: 1.00

---

Emotional Numbness

Reference — Score: 0.711

Model Responses

GPT-Realtime — Affection (1.52) | Target Score: 0.26

Gemini — Contemplation (1.52) | Target Score: 0.04

Grok — Affection (1.78) | Target Score: 0.00

---

Impatience and Irritability

Reference — Score: 1.056

Model Responses

GPT-Realtime — Amusement (1.02) | Target Score: 0.33

Gemini — Affection (1.29) | Target Score: -0.05

Grok — Astonishment (1.07) | Target Score: 0.54

---

Sadness

Reference — Score: 1.078

Model Responses

GPT-Realtime — Disappointment (1.33) | Target Score: 0.00

Gemini — Astonishment (1.13) | Target Score: 0.00

Grok — Contemplation (1.31) | Target Score: 0.95