Update README.md

README.md

@@ -69,6 +69,37 @@ The quality scores are distilled from two established audio quality models:
 | Content Enjoyment | 0.34 | 0.691 | 200 |
 
 
+### Group: Vocal, Demographic, and Psychological Attributes
+
+These 12 continuous dimensions capture the stylistic delivery, demographic traits, energetic state, and interpersonal tone of the speaker. They use varying scales to represent shifts from a neutral baseline.
+
+| Dimension | Scale Type | Range (Neutral) | Description |
+| --- | --- | --- | --- |
+| **Age** | Unipolar | 0 to 6 (-) | Estimated age from 0 (Infant/Toddler) to 6 (Very Old), e.g., 2=Teenager, 4=Adult. |
+| **Gender** | Bipolar | -2 to +2 (0) | Perceived gender representation, from Very Masculine (-2) to Very Feminine (+2). |
+| **Pitch (High vs. Low)** | Unipolar | 0 to 4 (2) | The fundamental frequency baseline, from Very High-Pitched (0) to Very Low-Pitched (4). |
+| **Talking Speed** | Custom | 0 to 36+ CPS (-) | The speech rate measured in Characters Per Second (CPS). |
+| **Valence** | Bipolar | -3 to +3 (0) | The overall positivity or negativity of the voice. |
+| **Arousal** | Unipolar | 0 to 4 (2) | The energy level of the speaker, ranging from very calm (0) to very excited (4). |
+| **Submissive vs. Dominant** | Bipolar | -3 to +3 (0) | The perceived authority or submission in the speaker's tone. |
+| **Serious vs. Humorous** | Unipolar | 0 to 4 (2) | The degree of humor or gravity in the delivery. |
+| **Confident vs. Hesitant** | Unipolar | 0 to 4 (2) | How certain or unsure the speaker sounds. |
+| **Soft vs. Harsh** | Bipolar | -2 to +2 (0) | The acoustic texture of the voice, from soothing and soft to abrasive and harsh. |
+| **Monotone vs. Expressive** | Unipolar | 0 to 4 (2) | The amount of dynamic pitch and rhythmic variation in the speech. |
+| **Warm vs. Cold** | Bipolar | -2 to +2 (0) | The perceived friendliness, empathy, or distance in the speaker's voice. |
+
+### Group: Emotional Categories
+
+Our dataset includes granular annotations for **40 distinct emotional categories** (e.g., *Joy, Anger, Gratitude, Sadness, Surprise*).
+
+> **Note on Sequence Length:** To manage sequence length and model focus, we condition **only on the top 5 most prominent emotions** present in any given audio sample.
+
+| Dimension Group | Scale Type | Range | Description |
+| --- | --- | --- | --- |
+| **40 Emotions** | Unipolar | 0 to 4 (Max 4.5) | Represents the intensity of a specific emotion. `0` means the emotion is completely absent, while `4+` indicates extreme manifestation. |
+
+
+
 ## How to Use
 
 ### Full Inference: Emotion Scores + Quality Scores