Datasets:

noty7gian
/

finetuned-hindi-punjabi-denoised

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Multilingual Speaker Diarization Dataset

This dataset contains synthetic multilingual speaker diarization data with Hindi, English, and Punjabi audio samples.

Dataset Structure

├── audio/           # WAV audio files (16kHz) - 627 files
├── csv/            # Individual CSV annotations - 627 files
├── rttm/           # RTTM format files for diarization - 627 files
├── all_samples_combined.csv  # Complete dataset annotations
└── all_samples_combined.rttm # Complete RTTM annotations

Statistics

Total samples: 627 audio files
Total duration: ~15 hours
Languages: Hindi, English, Punjabi (monolingual, bilingual, and trilingual conversations)
Speaker count: 2-5 speakers per conversation
Noise levels: Clean, low, medium, high noise conditions
Sample rate: 16kHz
Format: WAV audio files with CSV/RTTM annotations

Language Distribution

Hindi only: 83 samples (13.2%)
Punjabi only: 99 samples (15.8%)
English only: 64 samples (10.2%)
Bilingual: 189 samples (30.1%)
Trilingual: 192 samples (30.6%)

Usage

This dataset is designed for training and evaluating speaker diarization models, particularly for multilingual scenarios.

Loading the Dataset

# Load individual files
import pandas as pd
import librosa

# Load annotations
df = pd.read_csv("all_samples_combined.csv")

# Load audio file
audio, sr = librosa.load("audio/diarization_sample_0001.wav", sr=16000)

# Load RTTM annotations
with open("rttm/diarization_sample_0001.rttm", "r") as f:
    rttm_data = f.read()

File Format Details

CSV Format

AudioFileName: Name of the audio file
Speaker: Speaker ID (Speaker_00, Speaker_01, etc.)
StartTS: Start timestamp in seconds
EndTS: End timestamp in seconds
Language: Language code (hi, en, pa)

RTTM Format

Standard RTTM format for speaker diarization evaluation:

SPEAKER filename 1 start_time duration <NA> <NA> speaker_id <NA> <NA>

Citation

@dataset{multilingual_speaker_diarization_2025,
  title={Multilingual Speaker Diarization Dataset},
  author={Audio Agnostic Team},
  year={2025},
  url={https://huggingface.co/datasets/noty7gian/finetuned-hindi-punjabi-denoised}
}

License

This dataset is released under the MIT License.