datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
hinglish
Hinglish Concatenated 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, unified into a single Parquet dataset with consistent schema.
At a Glance
Stat
Value
Total clips
815,171
Total Estimated Hours
2,264+
Unique speakers
6,304
Raw audio size
~243 GB
Languages
Hindi (hi), Hinglish (hi-en), Indian… See the full description on the dataset page: https://huggingface.co/datasets/agarwalayushi/hinglish.MUCS-Hinglish
MUCS
Dataset Description
This dataset is a HuggingFace/Transformers compatible version of the MUCS 2021 Hinglish dataset.
This dataset is part of the MUltilingual and Code-Switching ASR Challenges for Low Resource Indian Languages challenge, subtask 2.
As this dataset is in Hinglish, it contains codeswitching between Hindi and English. The original dataset was found here.
In addition to making the dataset compatible for Transformers, preprocessing has been applied to… See the full description on the dataset page: https://huggingface.co/datasets/dianavdavidson/MUCS-Hinglish.S2R_Shrutilipi_hindi
Paytmlabs/S2R_Shrutilipi_hindi
Hindi speech dataset prepared from ai4bharat/Shrutilipi for Ultravox training.
Viewing samples on Hugging Face
The hindi config stores audio inside Parquet. The website dataset viewer often cannot decode that and shows no rows.
To inspect examples in the browser, open the Subset (config) drop-down and choose hindi_text_samples — text and continuation only (~2000 rows).
Ultravox training should keep using subset hindi (full audio).… See the full description on the dataset page: https://huggingface.co/datasets/Paytmlabs/S2R_Shrutilipi_hindi.hindi-audio-stories-20-30s
Hindi Audio Stories — 20–30 s clips (Qwen3-ASR, denoised)
Paired (audio, text) Hindi speech dataset. Each ~20–30 s denoised clip has
its transcript in two scripts, stored as separate rows (script column):
devanagari (Hindi) and latin (Hinglish romanization, uroman).
⚠️ Adult (NSFW) content. Research / non-commercial.
Stats
~12.6k clips × 2 scripts ≈ 25k rows · ~94 h · mean 26.8 s (97% in 20–30 s)
Audio: 24 kHz mono FLAC, UVR vocal-isolated (Mel-Band RoFormer —… See the full description on the dataset page: https://huggingface.co/datasets/backpropSukuna/hindi-audio-stories-20-30s.hinglish-casual
Hinglish Casual Speech
33,275 casual Hindi-English code-switched utterances (~31 GB) with audio,
transcripts in both Devanagari and Latin script (utterance /
utterance_latin), speaker ids, style metadata and durations. Full schema is in
the YAML header above.
Collected during the TinyAya programme to probe code-switched speech, which
neither the FLORES-derived text nor the TTS corpora cover. It is not part of
the v0.3 Stage-2 training set — that is
tr-hi-mimi-encoded.
from… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/hinglish-casual.hindi-english-asr
Hindi-English Code-Switching Conversational Audio
This dataset contains conversational Hindi-English code-switching audio recordings with human-verified transcripts. The conversations feature natural, spontaneous speech between multiple speakers who fluidly switch between Hindi and English, a common pattern in urban South Asian speech communities.
Transcription Process
All transcripts go through at least two full human review passes. First, a native-speaker… See the full description on the dataset page: https://huggingface.co/datasets/liva-ai/hindi-english-asr.mindbridge-phq9-hindi-audio-fixtures
MindBridge Hindi PHQ-9/GAD-7 — Audio Fixtures (30 clips)
Hindi audio fixtures for OIWER (Orthographically-Informed Word Error Rate,
AI4Bharat metric) audio-quality benchmarking on Gemma 4 E2B's native USM
conformer audio encoder. Used to verify post-fine-tune audio quality has
NOT regressed vs base E2B (audio modules explicitly frozen via
requires_grad=False during training to preserve the native USM encoder).
Recording setup
30 clips spanning PHQ-9 Sections A-D… See the full description on the dataset page: https://huggingface.co/datasets/Huzayfah-Patel/mindbridge-phq9-hindi-audio-fixtures.indian-english-hindi-tts-60min
Indian English + Hindi TTS Dataset
A small, heavily-curated Text-to-Speech corpus: 73.2 minutes (37.7 min Indian
English, 35.5 min Hindi) of single-speaker, studio-grade clips. Every clip's audio
was listened to and its transcript corrected against automated Sarvam ASR output;
resulting WER against the corrected text is 0.05% (en-IN) and 0.0% (hi-IN), showing
both very clean source audio and very accurate ASR. Built for the Sarvam AI ML &
Speech Data Pipeline assignment using a… See the full description on the dataset page: https://huggingface.co/datasets/auraCodes/indian-english-hindi-tts-60min.bharatvani-hindi-showcase
BharatVani Hindi Speech Corpus • Public Interactive Showcase
150-Hour Enterprise Devanagari Hindi Speech Corpus & Precomputed Latents
Curated & Mastered by BharatVani AI • TheCreatorOS
1. Interactive Dataset Preview
This repository is the official public evaluation showcase for the 150-Hour BharatVani Hindi Speech Corpus (103,784 Studio Clips).
Use the Dataset Viewer above to play real audio clips, inspect the word-level timestamp alignments, and… See the full description on the dataset page: https://huggingface.co/datasets/Sheeba2026/bharatvani-hindi-showcase.Hindi-speech-instruct
Hindi LLaMA-Omni Instruct Dataset
A Hindi speech instruction-following dataset designed for training speech-language models such as LLaMA-Omni. Each example pairs a spoken Hindi user question (audio) with a text assistant response.
Dataset Summary
Property
Value
Language
Hindi (hi)
Total examples
~110,718
Train split
~105,000 examples (batches 001–210)
Validation split
~5,500 examples (batches 211–222)
Audio format
FLAC, 16,000 Hz mono… See the full description on the dataset page: https://huggingface.co/datasets/Pastaaaaa2003/Hindi-speech-instruct.hindi_tts_dataset
Hindi TTS Dataset
This is a processed Hindi Text-to-Speech (TTS) dataset.
parakeet-hindi-asr
Parakeet Hindi-English Bilingual ASR
Fine-tuning NVIDIA Parakeet TDT 0.6B for bilingual Hindi-English automatic speech recognition.
Quick Start
# Download
pip install huggingface_hub
huggingface-cli download ketav/parakeet-hindi-asr --repo-type dataset --local-dir ./parakeet-hindi-asr
# Install dependencies
pip install nemo_toolkit[asr] bitsandbytes sentencepiece
# Train (after updating paths in config)
cd parakeet-hindi-asr/scripts
python ft_0.6B_hi_v3.py… See the full description on the dataset page: https://huggingface.co/datasets/ketav/parakeet-hindi-asr.vividh-test-hindi
🎙️ Vividh-ASR Benchmark — Hindi (Test Split)
How well does your ASR model actually work in the wild?
Vividh-ASR is a complexity-stratified benchmark that tells you exactly where your model succeeds — and where it falls apart.
Most Indic ASR benchmarks evaluate models on clean, studio-recorded speech. Real-world audio is not that. Vividh-ASR organises evaluation by acoustic complexity rather than domain, exposing the studio-bias that plagues models fine-tuned predominantly on read… See the full description on the dataset page: https://huggingface.co/datasets/adalat-ai/vividh-test-hindi.Hindi_Call_Center_Audio_Dataset_Dual_ChannelDataset Description:
This dataset is a large-scale collection of 1,587,658 hours of processed Hindi dual-channel call center audio recordings, containing 3,569,083 hours of processed call center audio recordings across 54 languages, designed to support the development and training of advanced speech AI and conversational AI systems.
It consists of real-world customer and agent speech recordings collected from call center environments. The dataset is organized in a dual-channel format, where… See the full description on the dataset page: https://huggingface.co/datasets/InfoBayAI/Hindi_Call_Center_Audio_Dataset_Dual_Channel.hindi-asr-benchmark
SkunkWorks Hindi STT Benchmark
Hindi ASR benchmark evaluating SkunkWorks alongside major commercial Hindi STT providers
(ElevenLabs, Deepgram, Sarvam) across 6 held-out evaluation subsets.
Subsets
config
source
n
kathbath
AI4Bharat Kathbath
1,929
kathbath_noisy
Kathbath noisy mic conditions
1,929
commonvoice
Mozilla Common Voice Hindi
1,727
mucs
MUCS 2021 Hindi subtask
3,897
fleurs
Google FLEURS hi_in test
418
indictts
AI4Bharat IndicTTS
100… See the full description on the dataset page: https://huggingface.co/datasets/SkunkWorkLabs/hindi-asr-benchmark.mucs-hinglish-blindtestHindi-Call-Center-Audio-Dataset-Single-ChannelDataset Description:
This dataset is a large-scale collection of 1,587,658 hours of processed Hindi single-channel call center audio recordings, containing 3,569,083 hours of processed call center audio recordings across 54 languages, designed to support the development and training of advanced speech AI and conversational AI systems.
The dataset captures authentic speech characteristics such as tone variation, pauses, silence patterns, interruptions, and natural speaking behaviour commonly… See the full description on the dataset page: https://huggingface.co/datasets/InfoBayAI/Hindi-Call-Center-Audio-Dataset-Single-Channel.hindi-youtube-asr-transcripts
Hindi YouTube ASR Transcripts
Auto-generated YouTube transcripts (VTT) from 21 Hindi channels for training ASR and TTS models.
Quick Start
# Download and extract
wget https://huggingface.co/datasets/ketav/hindi-youtube-asr-transcripts/resolve/main/youtube_asr_data.tar.gz
tar -xzf youtube_asr_data.tar.gz
Stats
Metric
Value
Channels
21
Total videos
109,981
Total hours
22,186.3
Hindi subtitles
98,309
Usable hours
19,032.5
Period
2025-2026… See the full description on the dataset page: https://huggingface.co/datasets/ketav/hindi-youtube-asr-transcripts.sarvam-indian-eng-hin-tts
Indian English + Hindi TTS Dataset (emotion-tagged)
A curated, single-speaker-per-clip speech dataset for Text-to-Speech research,
covering Indian English and Hindi. Every clip is sourced from YouTube,
transcribed with Sarvam Saaras v3, and emotion-tagged via acoustic cues + a Sarvam LLM.
Total: 82 clips, 55.6 minutes
Hindi: 28.8 min | Indian English: 26.8 min
Audio: mono, 24 kHz, 16-bit WAV
Single speaker per clip, clean (no background music / overlapping speakers)… See the full description on the dataset page: https://huggingface.co/datasets/Abhi29112005/sarvam-indian-eng-hin-tts.Indic_Hindi-English_Parallel_Speech
Dataset Access Information
This dataset is provided for research and academic purposes. Access to the dataset is gated, and users must request permission before downloading.
Dataset Summary
This repository contains the Hindi–English Speech-to-Speech Translation (S2ST) dataset introduced in the paper:
Benchmarking Hindi-to-English Direct Speech-to-Speech Translation with Synthetic Data
The dataset is designed to support research on direct speech-to-speech translation… See the full description on the dataset page: https://huggingface.co/datasets/mahendraphd/Indic_Hindi-English_Parallel_Speech.hindi-speech-recognition-dataset
Hindi Telephone Dialogues Dataset - 760 Hours
Dataset comprises 760 hours of high-quality audio recordings from 1,000+ native Hindi speakers, featuring telephone dialogues across diverse topics and domains. With a 95% sentence accuracy rate, this essential dataset is ideal for training and evaluating Hindi speech recognition systems. - Get the data
Dataset characteristics:
Characteristic
Data
Description
Audio of telephone dialogues in Hindi for training… See the full description on the dataset page: https://huggingface.co/datasets/ud-nlp/hindi-speech-recognition-dataset.Hindi-Speech-Dataset
🎧 Hindi Speech Dataset
The Hindi Speech Dataset is a high-quality and structured speech audio dataset developed to support modern AI systems that rely on diverse audio data and scalable voice data. It contains 132 hours of recordings distributed across 565 files, available in MP3 and WAV formats, with a total size of 101 MB. This carefully curated audio dataset provides balanced speaker representation with 49% female and 51% male contributors, covering an age range from 18 to 50+… See the full description on the dataset page: https://huggingface.co/datasets/Speech-data/Hindi-Speech-Dataset.hindi-english-codeswitch-dataset
Hindi-English Code-Switch ASR Transcripts
Text transcripts and metadata for a large bilingual Hindi-English code-switch ASR training corpus, used to train Abhisingh-18/hindi-english-codeswitch-asr.
This release contains transcripts and metadata only — no audio files. Audio was sourced from multiple corpora and institutions and is not redistributed here.
Credits
Speech data collection and curation credit: SPRING Lab, IIT Madras.
Contents
File… See the full description on the dataset page: https://huggingface.co/datasets/Abhisingh-18/hindi-english-codeswitch-dataset.hindi-speech-recognition-dataset
Hindi Speech Dataset for recognition task
Dataset comprises 760 hours of telephone dialogues in Hindi, collected from 1,000+ native speakers across various topics and domains. This dataset boasts an impressive 95% sentence accuracy rate, making it a valuable resource for advancing speech recognition technology.
By utilizing this dataset, researchers and developers can advance their understanding and capabilities in automatic speech recognition (ASR) systems, transcribing audio… See the full description on the dataset page: https://huggingface.co/datasets/UniDataPro/hindi-speech-recognition-dataset.cs-fleurs-hineng-read-test
CS-FLEURS Hindi-English Read Test
n = 233 utterances · private eval set for code-switched ASR benchmarking
Source
Derived from byan/cs-fleurs (read/test split, language = hin-eng).
CS-FLEURS is a code-switching extension of FLEURS covering Hindi-English mixed speech.
Construction
Filtered to language == "hin-eng" rows from the read/test split:
Audio downloaded directly from the byan/cs-fleurs HF repository
transcription = reference text (mixed… See the full description on the dataset page: https://huggingface.co/datasets/Trelis/cs-fleurs-hineng-read-test.Hindi_Podcast_Audio_Dataset_Dual_Channel
Dataset Description
This dataset is a large-scale collection of 11,607 hours of processed Hindi dual-channel podcast audio recordings, containing 57,569 hours of processed podcast audio recordings across 12 languages, designed to support the development and training of advanced speech AI and conversational AI systems.
It captures real-world podcast conversations across diverse topics and formats. The dataset is organized in a dual-channel format, where corresponding speaker… See the full description on the dataset page: https://huggingface.co/datasets/InfoBayAI/Hindi_Podcast_Audio_Dataset_Dual_Channel.hinglish-code-switched-conversations-v1
Hinglish Code-Switched Conversational Dataset v1
Overview
This dataset contains structured Hinglish conversational voice data built to reflect how people actually speak in real-world interactions.
Most speech datasets are clean, scripted, or heavily processed. That works in controlled testing, but it breaks in production where speakers interrupt each other, switch languages, use regional accents, pause mid-thought, and shift context naturally.
This sample release… See the full description on the dataset page: https://huggingface.co/datasets/sonexis-ai/hinglish-code-switched-conversations-v1.cleaned-asr-transcripts-hinglish
cleaned-asr-transcripts-hinglish
bingbangboom/cleaned-asr-transcripts-hinglish is a parallel corpus containing 14k+ pairs of raw-synthetic Hindi ASR (Automatic Speech Recognition) transcripts mapped to their clean, properly punctuated, and transliterated "Hinglish" (Romanized Hindi) counterparts.
This dataset is specifically designed for ASR post-processing, transliteration models, and fine-tuning Large Language Models (LLMs) to understand and generate high-quality, conversational… See the full description on the dataset page: https://huggingface.co/datasets/bingbangboom/cleaned-asr-transcripts-hinglish.cleaned-asr-transcripts-hinglish
cleaned-asr-transcripts-hinglish
bingbangboom/cleaned-asr-transcripts-hinglish is a parallel corpus containing 14k+ pairs of raw-synthetic Hindi ASR (Automatic Speech Recognition) transcripts mapped to their clean, properly punctuated, and transliterated "Hinglish" (Romanized Hindi) counterparts.
This dataset is specifically designed for ASR post-processing, transliteration models, and fine-tuning Large Language Models (LLMs) to understand and generate high-quality, conversational… See the full description on the dataset page: https://huggingface.co/datasets/SaiyanSai/cleaned-asr-transcripts-hinglish.hindi_and_english_stt_tts_master_data
Hindi and English STT/TTS Master Data
Combined speech dataset for Hindi and Indian English automatic speech recognition (ASR) and text-to-speech (TTS) training. Parquet shards embed WAV audio bytes with transcripts.
Dataset structure
hindi/<source>/train-*.parquet
english/<source>/train-*.parquet
Each config loads one source independently (~3.24M total rows, ~1.9 TB).
Features
Column
Type
Description
audio
Audio
WAV bytes embedded in… See the full description on the dataset page: https://huggingface.co/datasets/sajalmadan0909/hindi_and_english_stt_tts_master_data.