EmoTa is released by Team EmoTa / aaivu under the EmoTa Academic-Commercial License (EACL) v1.0 (academic use governed by CC BY-NC 4.0). Access is granted automatically once you agree to the terms below. By requesting access you confirm that you will use the dataset only for non-commercial academic research and education, that you will cite the EmoTa paper (CHiPSAL 2025), and that you will not redistribute, sell, sublicense, or publicly share the dataset in original or modified form without prior written permission from the Licensor. Commercial use requires a separate paid license — contact rtuthaya@cse.mrt.ac.lk. See LICENSE.txt for the full terms.

EmoTa: A Tamil Emotional Speech Dataset

EmoTa is the first emotional speech dataset in Tamil, designed to reflect the linguistic diversity of Sri Lankan Tamil speakers. It contains 936 recorded utterances from 22 native Tamil speakers (11 male, 11 female), each articulating 19 semantically neutral sentences across five emotions: anger, happiness, sadness, fear, and neutral.

🌐 Project page: https://aaivu.github.io/EmoTa/
💻 Code & loader: https://github.com/aaivu/EmoTa
📄 Paper (CHiPSAL 2025): https://aclanthology.org/2025.chipsal-1.19/
🏛️ Developed by Team EmoTa / aaivu, University of Moratuwa

⚠️ Gated, non-commercial dataset. Access is auto-approved after you accept the EmoTa Academic-Commercial License (EACL). Non-commercial academic use only.

Dataset at a glance

Property	Value
Utterances	936
Speakers	22 (11 male, 11 female)
Sentences	19 (semantically neutral)
Emotions	5 — angry, happy, sad, fear, neutral
Language	Tamil (`ta`), Sri Lankan Tamil dialects
Audio format	Mono WAV, 48 kHz, 16-bit PCM
Total duration	~43 minutes
Recording	Controlled, soundproof environment, professional equipment
Inter-annotator agreement	Fleiss' Kappa = 0.74 (substantial)

Emotion distribution

Emotion	Utterances
happy	209
neutral	209
sad	209
angry	199
fear	110
Total	936

Speaker demographics

Speakers span four provinces of Sri Lanka, capturing northern, eastern, western, and central Tamil dialects. Ages range from 22 to 27.

Province	Speakers
eastern	10
northern	7
western	4
central	1

Full per-speaker metadata (id, age, gender, region) is in meta/speaker_details.csv.

Data fields

Each row in data/metadata.csv describes one audio file:

Field	Description
`file_name`	Path to the WAV file (relative to `data/`)
`audio`	Decoded audio (auto-loaded by 🤗 Datasets)
`speaker_id`	Speaker identifier (1–22)
`gender`	Speaker gender (`male` / `female`)
`age`	Speaker age
`region`	Speaker's province (northern / eastern / western / central)
`sentence_id`	Sentence identifier (1–19)
`emotion`	Emotion label: `angry`, `happy`, `sad`, `fear`, `neutral`
`transcript`	Tamil transcript of the utterance
`duration_sec`	Clip duration in seconds

File naming

Audio filenames follow: <speakerID>_<sentenceID>_<emotion[:3]>.wav — e.g. 01_05_ang.wav = speaker 1, sentence 5, angry. Emotion prefixes: ang→angry, hap→happy, sad→sad, fea→fear, neu→neutral.

The 19 sentences

ID	Transcript (Tamil)
1	நான் இன்று மாலை வீட்டுக்கு செல்கிறேன்
2	இண்டைக்கு மழையா இருக்கு
3	நாங்க நல்லபடியாக செய்துமுடித்துள்ளோம்
4	எப்போதும் தாமதமாக வராதீர்கள்
5	நான் உன்னை காதலிக்கிறேன் அன்பே
6	அந்த செய்தித்தாளை இங்கு வையுங்கள்
7	இந்த நோயாளியின் உடல்நிலை எப்படி இருக்கிறது ?
8	இப்ப உனக்கு என்ன பிரச்சனை ?
9	உனக்கு யாரை ரொம்ப பிடிக்கும்?
10	என் பையை திருப்பிக் கொடு.
11	எல்லோரும் தவறு செய்கிறார்கள்.
12	நீ இப்போது வளர்ந்துவிட்டாய்.
13	நான் அதை பார்த்து கொள்கிறேன்.
14	நீங்கள் எங்கு போகிறீர்கள்?
15	புத்தகம் மேசையில் உள்ளது.
16	ரயில் மாலை 5 மணிக்கு வரும்.
17	எனக்கு வழி தெரியவில்லை.
18	நான் உன்னை சந்திக்க வேண்டும்.
19	அண்ணா எழுந்திருங்கள்

Usage

After you have been granted access, authenticate with huggingface-cli login (or set HF_TOKEN), then:

With 🤗 Datasets

from datasets import load_dataset

ds = load_dataset("aaivu-labs/EmoTa", split="train")
print(ds[0])
# {'audio': {...}, 'speaker_id': 1, 'gender': 'male', 'emotion': 'angry',
#  'transcript': 'நான் இன்று மாலை வீட்டுக்கு செல்கிறேன்', ...}

With the official `emota_loader`

pip install emota_loader

from emota_loader import EmoTaDataset

dataset = EmoTaDataset(root_dir="path/to/wav/files")
print(len(dataset))          # 936
sample = dataset[0]
print(sample.emotion, sample.speaker_gender, sample.transcript)

Baseline results

Initial evaluations for Speech Emotion Recognition (SER) reported in the paper:

Model	F1-score
XGBoost	0.91
Random Forest	0.90

License & access

This dataset is distributed under the EmoTa Academic-Commercial License (EACL) v1.0 (academic use governed by CC BY-NC 4.0). See LICENSE.txt for the full terms. Key points:

✅ Academic / research / educational use permitted, with attribution.
🚫 No commercial use without a separate paid license.
🚫 No redistribution, resale, sublicensing, or public sharing of the dataset (original or modified) without prior written permission from the Licensor.
📌 Access on the Hub is gated with automatic approval — you only need to accept the terms above.

For commercial licensing or any inquiry: rtuthaya@cse.mrt.ac.lk.

Citation

If you use EmoTa, please cite:

@inproceedings{thevakumar-etal-2025-emota,
    title = "{E}mo{T}a: A {T}amil Emotional Speech Dataset",
    author = "Thevakumar, Jubeerathan  and
      Thavarasa, Luxshan  and
      Sivatheepan, Thanikan  and
      Kugarajah, Sajeev  and
      Thayasivam, Uthayasanker",
    editor = "Sarveswaran, Kengatharaiyer  and
      Vaidya, Ashwini  and
      Krishna Bal, Bal  and
      Shams, Sana  and
      Thapa, Surendrabikram",
    booktitle = "Proceedings of the First Workshop on Challenges in Processing South Asian Languages (CHiPSAL 2025)",
    month = jan,
    year = "2025",
    address = "Abu Dhabi, UAE",
    publisher = "International Committee on Computational Linguistics",
    url = "https://aclanthology.org/2025.chipsal-1.19/",
    pages = "193--201",
    abstract = "This paper introduces EmoTa, the first emotional speech dataset in Tamil, designed to reflect the linguistic diversity of Sri Lankan Tamil speakers. EmoTa comprises 936 recorded utterances from 22 native Tamil speakers (11 male, 11 female), each articulating 19 semantically neutral sentences across five primary emotions: anger, happiness, sadness, fear, and neutrality. To ensure quality, inter-annotator agreement was assessed using Fleiss' Kappa, resulting in a substantial agreement score of 0.74. Initial evaluations using machine learning models, including XGBoost and Random Forest, yielded a high F1-score of 0.91 and 0.90 for emotion classification tasks. By releasing EmoTa, we aim to encourage further exploration of Tamil language processing and the development of innovative models for Tamil Speech Emotion Recognition."
}

Team

Developed by Team EmoTa at aaivu, University of Moratuwa, Sri Lanka:

Jubeerathan Thevakumar
Luxshan Thavarasa
Thanikan Sivatheepan
Sajeev Kugarajah
Uthayasanker Thayasivam

Acknowledgements

We thank all 22 volunteer speakers and the annotators who contributed to EmoTa.

Downloads last month: 13

Total file size:

376 MB

Access EmoTa — Academic, Non-Commercial Use Only