Datasets:
Dataset Viewer
The dataset viewer is not available for this dataset.
Unexpected token '<', "<html>
<h"... is not valid JSON
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Indian TTS Dataset — Hindi + Indian English
A curated, single-speaker TTS training dataset with 144 segments (~60 minutes total) sourced from YouTube, transcribed using Sarvam AI ASR, and annotated with emotion/style tags.
Dataset Summary
| Split | Segments | Duration |
|---|---|---|
| Indian English | 71 | 29.6 min |
| Hindi | 73 | 30.8 min |
| Total | 144 | 60.3 min |
Audio Specs
- Sample rate: 16 kHz
- Channels: Mono
- Format: WAV (16-bit PCM)
- Segment length: 20–28 seconds
Columns
| Column | Description |
|---|---|
audio |
Audio segment (16kHz mono WAV) |
transcript |
ASR transcript (Sarvam saarika:v2.5) |
language |
indian_english or hindi |
language_code |
en-IN or hi-IN |
emotion_tag |
Emotion/style label |
source_channel |
YouTube channel name |
style |
Content style (news_formal, conversational_podcast, etc.) |
duration_s |
Segment duration in seconds |
mean_dbfs |
Mean audio level in dBFS |
Emotion Distribution
| Emotion | Count |
|---|---|
| questioning | 56 |
| formal | 27 |
| informational | 19 |
| motivational | 18 |
| conversational | 17 |
| emphatic | 2 |
| excited | 2 |
| storytelling | 1 |
| angry | 1 |
| sad | 1 |
Sources
Indian English: The Seen and the Unseen
Hindi: Aaj Tak, Dhruv Rathee, Josh Talks Hindi
Pipeline
- Audio downloaded via
yt-dlp, converted to 16kHz mono WAV viaffmpeg - Silence-based segmentation into 20–28s chunks
- Transcription via Sarvam AI
saarika:v2.5ASR - Automated QC: SNR check, transcript length, ASR artifact detection
- Emotion tagging: rule-based classifier on transcript + source style
See the full pipeline code at: https://github.com/itsharshi/indian-tts-dataset
License
Audio sourced from YouTube under fair use for research/educational purposes. Dataset annotations (transcripts, tags) are released under CC BY 4.0.
- Downloads last month
- 13