Loading datasets...
Release Date: 11/5/2026
Format: TSV, WAV
Size: 13.57 GB
This dataset contains 68 hours of read Saraiki speech recorded for text-to-speech development: 19,892 clips from three native speakers from South Punjab, Pakistan (Spk-04, male, 40–49, Khanewal, 5.8 h; Spk-05, female, 20–29, Kabirwala, 36.6 h; Spk-06, female, 30–39, Multan, 25.6 h). Each clip is paired with its sentence in Saraiki Perso-Arabic script, including the Saraiki letters ٻ ݙ ݨ. Sentences were sourced from Kaleem Arts Press and cover 17 domains, including literature, news, education, religion, storytelling, poetry, formal and informal Saraiki, and children’s content. Audio is 44.1 kHz, 16-bit mono WAV, recorded with a Boya K3 microphone in a studio setup using the SABRe recording app, with corpus-wide SNR of about 61–66 dB. Recordings are organised in session folders by speaker and domain, each with a mapping.tsv linking audio files to sentences. All speakers gave written consent and were paid. The data is part of MirasAI’s 200-hour Saraiki TTS corpus and is licensed for commercial use and model training under the MirasAI Dataset Licence.
Pricing details
Your purchase is the license to the raw data. Once purchased, you're responsible for storing and using this dataset.
Your purchase includes a license to the data, paid directly to the dataset vendor, and a 5% ($340.00) platform fee paid to MDC.
Read More About The PoliciesLicensed under the MirasAI Dataset Licence (https://mirasai.net/license): commercial use and model training are permitted, but the audio may not be redistributed, resold or sublicensed, used to build competing datasets, or used to identify speakers.
The buyer agrees to abide by the license terms.
Ethical Review
This is a commercial data collection by MirasAI LLC, not an academic study, so no review board was involved. Each speaker was told before recording that the data was for a Saraiki text-to-speech dataset, would be licensed commercially including for training speech synthesis models that reproduce their voice, and would be distributed through the Mozilla Data Collective. Each speaker signed a consent form covering these uses and was paid. Consent forms are available on request. Speakers are identified by ID only, with gender, age band and region as the only published metadata. The audio is read speech of supplied sentences, so it contains no personal information. The licence prohibits identifying speakers or redistributing the audio.
Intended Use
Speech to Text for Saraiki.
