Datasetpublic

Srijan-Upadhyay/marathi-tts-yt-trimmed

Marathi TTS YouTube Trimmed Audio Corpus Dataset Summary Marathi TTS YouTube Trimmed Audio Corpus comprises preprocessed, segmented, and silence-trimmed Marathi speech audio clips derived from curated public Marathi educational, monologue, and conversational videos. Each clip is carefully aligned with clean Marathi text transcriptions for high-clarity speech model training. Dataset Structure Format: Segmented audio files (WAV, 22.05 kHz) + text… See the full description on the dataset page: https://huggingface.co/datasets/Srijan-Upadhyay/marathi-tts-yt-trimmed.

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes15downloads

Dataset Card

Marathi TTS YouTube Trimmed Audio Corpus

Dataset Summary

Marathi TTS YouTube Trimmed Audio Corpus comprises preprocessed, segmented, and silence-trimmed Marathi speech audio clips derived from curated public Marathi educational, monologue, and conversational videos. Each clip is carefully aligned with clean Marathi text transcriptions for high-clarity speech model training.

Dataset Structure

—Format: Segmented audio files (WAV, 22.05 kHz) + text transcripts.
—Language: Marathi (mr)
—Preprocessing: Silence removal, loudness normalization, noise filtering, and text alignment.

Data Fields

—id: Unique audio segment ID.
—audio_path: Path to the trimmed audio file.
—text: Clean Devanagari text matching the spoken audio.
—duration_sec: Duration of the segment (typically 2-10 seconds).

Usage

python

from datasets import load_dataset

ds = load_dataset("Srijan-Upadhyay/marathi-tts-yt-trimmed", token=True)
print(ds)

Intended Use

For voice cloning, TTS model dataset expansion, and speech recognition benchmarking in Marathi.