AI4SER: Artificial Intelligence for Speech Emotion Recognition

AI4SER is an emotional speech dataset designed for Speech Emotion Recognition (SER) tasks.
It contains utterances in Italian, recorded by multiple speakers, and annotated with 7 emotional categories.

The dataset has been created following Open Science and FAIR data principles, and it is publicly available under a CC BY 4.0 license.

---

🧠 Emotion Labels

The emotional category of each utterance is encoded in the filename using a 3-letter prefix:

Code	Emotion
dis	Disgust
hap	Happiness
fea	Fear
ang	Anger
sur	Surprise
sad	Sadness
neu	Neutral

---

🗂️ File Naming Convention

Each .wav file is named using the following format:

<emotion_code>_<phrase_id>_<speaker_id>.wav

For example:

• hap_03_07.wav: Happiness, phrase 03, speaker 07
• sad_10_02.wav: Sadness, phrase 10, speaker 02

---

📁 Directory Structure

The dataset is organized as follows:

 
AI4SER/
├── 01/ # Speaker 01
│ ├── dis_01_01.wav
│ ├── hap_02_01.wav
│ └── ...
├── 02/ # Speaker 02
│ ├── ang_03_02.wav
│ ├── neu_04_02.wav
│ └── ...
├── ...

Each subfolder corresponds to a speaker ID.

---

📐 Technical Details

• Format: 16-bit PCM WAV
• Sampling Rate: 44,100 Hz
• Language: Italian
• Utterances per speaker: 70 (7 emotions × 10 phrases)
• Duration: Variable

---

🏷️ Annotations

In addition to the main dataset splits, we provide four supplementary CSV files (values are separated by ;):

File name	Description
EMOTIONS-HIT-RATIO.csv	Fraction of listeners who identified each audio file with one of the 7 considered discrete emotions.
VALIDATION.csv	Mean value of the Validation parameter assigned by listeners for each audio file.
DOMINANCE.csv	Mean value of the Dominance parameter assigned by listeners for each audio file.
AROUSAL.csv	Mean value of the Arousal parameter assigned by listeners for each audio file.

Rating scales

Listeners annotated valence, dominance, and arousal using discrete attributes mapped onto the following numerical scale:

Value	Valence interpretation	Dominance interpretation	Arousal interpretation
-3	Very negative	Very submissive	Very calm
-2	Negative	Submissive	Calm
-1	Slightly negative	Slightly submissive	Slightly calm
0	Neutral	Neutral	Neutral
+1	Slightly positive	Slightly dominant	Slightly intense
+2	Positive	Dominant	Intense
+3	Very positive	Very dominant	Very intense

---

📄 License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

---

📚 Citation & Funding Acknowledgement

This dataset was developed by the Digital Signal Processing Lab – Department of Engineering – University of Messina.

This work was supported by the European Union – Next Generation EU under the Italian National Recovery and Resilience Plan (NRRP), Mission 4, Component 2, Investment 1.3, CUP C49J24000240004, partnership on “Telecommunications of the Future” (PE00000001 – program “RESTART”).

If you use this dataset in your research, please cite it appropriately (a formal citation will be added upon publication).

---

📦 Usage with Hugging Face 🤗 Datasets

Once published, you can load the dataset using:

from datasets import load_dataset

dataset = load_dataset("your-username/AI4SER")