Hinglish Audio Dataset

A large-scale, cleaned and annotated speech dataset covering Hindi, Hinglish (Hindi–English code-switching), and Indian English — compiled from 14 public corpora and original custom recordings.

The full dataset (Parquet + audio) is hosted on Hugging Face: 👉 huggingface.co/datasets/agarwalayushi/hinglish

---

Overview

Stat	Value
Total clips	815,171
Total duration	~2,264 hours
Unique speakers	6,304
Raw audio size	~237 GB
Languages	Hindi · Hinglish · Indian English
Audio format	WAV (native sample rates)
Dataset format	Parquet with embedded audio (HF-compatible)
Primary tasks	ASR · TTS fine-tuning · Voice cloning

---

Dataset Schema

Column	Type	Description
source	string	Speaker or source identifier
text	string	Cleaned transcript (includes <hi-en> tags for code-switched utterances)
audio	Audio	Waveform bytes + sample rate, embedded in Parquet
quality	string	Sample rate (Hz) or MOS quality score where available
duration	string	Clip duration in seconds

---

Source Datasets

This dataset aggregates the following public corpora alongside original custom recordings:

Source	Clips	Description	License
NPTEL Hindi Spoken Tutorial (AI4Bharat)	521,028	Educational lectures in Hindi	CC BY 4.0
AI4Bharat Kathbath	94,903	Multi-speaker Hindi speech benchmark	CC BY 4.0
AI4Bharat IndicTTS	36,613	Studio-quality Hindi TTS corpus	CC BY 4.0
Hinglish — ujs	25,378	Code-switched Hindi–English speech	See source
Mozilla Common Voice 17 (Hindi)	24,643	Community-contributed Hindi speech	CC0 1.0
AI4Bharat Mann Ki Baat	22,483	Hindi broadcast speech, En-Indic aligned	CC BY 4.0
Mann Ki Baat (English)	22,477	English side of Mann Ki Baat parallel corpus	CC BY 4.0
Hindi Female Single Speaker HQ — Shekharmeena	22,058	High-quality single female speaker Hindi	See source
Orpheus TTS Indian English — ar17to	18,238	Multi-speaker Indian English TTS	See source
Indic TTS Hindi — SPRINGLab	11,825	Multi-speaker Hindi TTS	CC BY 4.0
Indian English — krishan23	6,765	Indian-accented English speech	See source
Hinglish Test TTS — Shekharmeena	3,136	Custom Hinglish TTS recordings	See source
Custom female TTS recordings	2,843	Original studio recordings	CC BY 4.0
Orpheus TTS Shaurya — prashantarya	1,419	Male Hindi/English TTS voice	See source
Anika Voice — Shekharmeena	~5	Custom female Hindi/Hinglish voice	See source

---

Curation & Processing

• All clips re-segmented to remove silence and cross-talk
• Transcripts normalised to Unicode NFC; <hi-en> language tags added for code-switched utterances
• Duplicate and near-duplicate clips removed across sources
• Consistent 5-column schema applied before concatenation (source, text, audio, quality, duration)
• Audio stored as WAV; original sample rates preserved per speaker/source

---

Usage

Load from Hugging Face

from datasets import load_dataset

ds = load_dataset("agarwalayushi/hinglish", split="train", streaming=True)

for sample in ds:
    audio = sample["audio"]        # {"array": ..., "sampling_rate": ...}
    text  = sample["text"]
    print(text, audio["sampling_rate"])

Filter by language

# Code-switched Hinglish utterances only
hi_en = ds.filter(lambda x: x["text"].startswith("<hi-en>"))

---

License

Released under CC BY 4.0.

• ✅ Free for research and commercial use
• ✅ Redistribution and adaptation allowed
• 📌 Attribution required — cite this dataset and link to upstream sources
• 🚫 Do not use to generate non-consensual synthetic voices of real individuals

Constituent datasets retain their original licenses. Users are responsible for compliance with each upstream source's terms.

---

Citation

@dataset{agarwal2026hinglish,
  author    = {Agarwal, Ayushi},
  title     = {Hinglish Concatenated Audio Dataset},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/agarwalayushi/hinglish},
  note      = {Aggregated from Mozilla Common Voice, AI4Bharat (Kathbath, Mann Ki Baat,
               Spoken Tutorial, IndicTTS), SPRINGLab IndicTTS, and custom recordings}
}

---

Contributing & Issues

To report transcription errors, missing attributions, or audio quality issues, please open an issue or start a discussion on Hugging Face.