humyn-labs/Indian-Emotional-Speech-Corpus

Indian Emotional Speech Corpus

Dataset Description

This dataset comprises high-quality audio recordings of Indian speakers reading a standardized 50-word paragraph in four distinct emotional tones — happy, sad, surprised, and angry.

Each recording is approximately 20–25 seconds long and includes the full paragraph with tone shifts at specific points.

Text spoken by all participants:

(happy tone) Last Monday was perfect—I got the job I’d been dreaming of! I screamed, hugged my parents, and we even had cake for breakfast. (sad tone) But two days later, my mom called—our dog Bruno had passed away in his sleep. I didn’t even get to say goodbye. (surprised tone) And just when I thought the week couldn’t get any crazier, I got a surprise call—I was selected for a global fellowship I never even applied for! (angry tone) But earlier that day, my Wi-Fi cut out again during an important meeting... third time this week! It’s beyond frustrating. What a week. Honestly, I’m still processing it all.

The dataset supports training and evaluation of models in:

• —Automatic Speech Recognition (ASR)
• —Emotional tone classification
• —Voice synthesis and generation
• —Emotion-aware conversational agents

Dataset Structure

• —Audio Files: Audio clips are stored in the audio/ directory with names like audio_001.wav, audio_002.mp3, etc.
• —Formats: .wav, .mp3, .m4a
• —Duration: Each file is 20–25 seconds long

• —Metadata (in metadata.csv):
• —file_name: Name of the audio file
• —transcription: Full 50-word paragraph
• —emotional_tones: Order of tones (e.g., happy;sad;surprised;angry)
• —age_group: e.g., 18–30, 31–45, 46–60
• —gender: male or female
• —region: e.g., North India, South India, East India, West India

Intended Uses

✅ Direct Use

• —Training and benchmarking ASR models with Indian-accented English
• —Emotion detection and classification from voice
• —Research in affective computing and empathetic AI

❌ Out-of-Scope Use

• —Real-time or production-grade systems
• —Commercial use without proper CC BY 4.0 attribution
• —Clinical or diagnostic use cases

Considerations and Limitations

• —❗ The dataset is small (<1,000 samples) and not fully representative of India's linguistic and emotional diversity
• —💡 Emotions are subjective — classification results may vary by listener or model
• —🔄 Future versions will aim to expand multilingual support and speaker diversity

License

CC BY 4.0 — You can use, modify, and share the dataset with appropriate credit.

Contact

• —For queries or collaborations related to datasets, contact at :
• —support@humynlabs.ai ---

Citation

BibTeX:

@misc{indian_emotional_speech_corpus,
  title = {Indian Emotional Speech Corpus},
  author = {Contributors from across India},
  year = {2025},
  howpublished = {\url{https://huggingface.co/datasets/indian_emotional_speech_corpus}},
  note = {Dataset available under CC-BY-4.0 License}
}