---
license: cc-by-4.0
task_categories:
- automatic-speech-recognition
tags:
- audio
- indic
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: file_name
    dtype: string
  - name: language
    dtype: string
  splits:
  - name: train
    num_bytes: 8169163578.0
    num_examples: 93
  download_size: 5351310711
  dataset_size: 8169163578.0
---
# Indic Conversational ASR Dataset

## 📌 Dataset Overview

This dataset contains high-quality audio samples curated for **Automatic Speech Recognition (ASR)** tasks.  

The recordings are optimized for speech recognition and are provided with:

- **Sampling Rate:** 16 kHz – 24 kHz  
- **Bit Depth:** 16-bit  
- **Audio Type:** Non-scripted conversational speech  
- **Format:** Dual-speaker conversations  

---

## 🌏 Supported Languages & Variants

| | | | | |
|---|---|---|---|---|
| Telugu | Kannada | Malayalam | Bengali (IN – Kolkata) | Bengali (IN – Non-Kolkata) |
| Bengali (BD)| Assamese | Odia | Gujarati | Marathi | Punjabi | Bhojpuri | Haryanvi |
| Tamil | Tamilish | Hinglish | Marvadi | Chhattisgarhi |

---

### 👥 Speaker Representation

- Dual-speaker conversational recordings  
- Natural, spontaneous speech  
- Female speaker representation: **~20%–30%**

---

# 🛠 Dataset Creation Methodology

## 📥 Data Collection

Speech data was collected through micro-communities across India and neighboring regions, spanning:

- Tier 1 cities  
- Tier 2 cities  
- Tier 3 cities  

This approach ensured:

- Linguistic diversity  
- Regional accent coverage  
- Authentic conversational patterns  

---

## 🎙 Recording Setup

- Non-scripted, dual-speaker conversations  
- Duration: **10–30 minutes per recording**  
- Topics include:
  - Business
  - Finance
  - Politics
  - Daily-life discussions

---

## ✅ Quality Validation

All audio samples underwent automated and manual validation.

### 🔍 Automated Checks
SRMR • SIGMOS • VQScore • WVMOS  

**Evaluated:** Signal quality, perceptual clarity, speech intelligibility.

### 👥 Human Review
Ensured conversational naturalness, audio clarity, and ASR suitability.
 

---

# 🎯 Dataset Intended Purpose

## ✔️ Intended Uses

This dataset is designed for:

- Training and fine-tuning **Automatic Speech Recognition (ASR)** models  
- Benchmarking conversational ASR systems  
- Code-mixed speech recognition research  
- Speaker turn detection and interruption modeling  
- Informal and spontaneous speech modeling  
- Emotion recognition research  
- Speaker interaction analysis  
- Conversational AI research  
- Academic and open-source research for low-resource Indic languages  

---

## 🚫 Out-of-Scope Uses

This dataset is **not intended for**:

- Real-time, safety-critical, or production-grade systems without additional validation  
- Commercial deployment without proper attribution and compliance with **CC BY 4.0**  
- Medical, clinical, legal, or diagnostic decision-making applications  

---

# 📜 License

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

# 📬 Contact

For queries regarding this dataset, please reach out to:

**[arunabh@humynlabs.ai]**

---