# Speech Recognition Datasets: AI Datasets Source: https://savrn.com/datasets/tasks/speech-recognition Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Datasets by Task # Speech Recognition Datasets 35 open-weight speech recognition datasets in the SAVRN Model Hub, with Shrikant Nayak, The Larry Project and Thant Zin Phyo publishing the most. 35Datasets 26Publishers 8Licenses ## SAVRN's Take The speech recognition datasets here are few but telling. Google's FLEURS, which covers about a hundred languages with the same sentences read aloud, is the standard multilingual benchmark. The rest are mostly Indian-language collections, including call-center speech and university-contributed recordings, which reflect how much open speech work is now aimed at underserved languages. Most of this category is CC-BY-4.0, so commercial use is allowed with attribution. Speech data carries extra obligations beyond the license: speakers' voices are personal data in many jurisdictions. Check how consent was collected before training on recordings of real people, especially call-center audio. SAVRN Research, 2026-09-18 ## Most Downloaded | Dataset | Publisher | License | Monthly downloads | | --- | --- | --- | --- | | [fleurs](https://savrn.com/datasets/fleurs) | Google | cc-by-4.0 | 102.1k | | [Quranic-Recitation-Data](https://savrn.com/datasets/quranic-recitation-data) | Muhammad Zohaib Hassan | apache-2.0 | 19.4k | | [balalaika-sft-rover7-punct](https://savrn.com/datasets/balalaika-sft-rover7-punct) | Lab260 | Not stated | 18.6k | | [bolAIndia](https://savrn.com/datasets/bolaindia) | Kapture CX | Not stated | 18k | | [QuranLab-Arabic-Speech](https://savrn.com/datasets/quranlab-arabic-speech) | Mostafa Mahdi | other | 3.9k | | [unnamed-split-ar](https://savrn.com/datasets/unnamed-split-ar) | Ahmed Wasfy | cc-by-3.0 | 3.6k | | [dialectal-arabic-voices](https://savrn.com/datasets/dialectal-arabic-voices) | Moaead | Not stated | 1.7k | | [Vartalaap](https://savrn.com/datasets/vartalaap) | Kapture CX | cc-by-nc-4.0 | 1.7k | | [ralomi-reazonspeech-20h-development](https://savrn.com/datasets/ralomi-reazonspeech-20h-development) | TTS Dataset | cdla-sharing-1.0 | 1.6k | | [DNU-ViSpeech](https://savrn.com/datasets/dnu-vispeech) | Trang Lê | cc-by-nc-4.0 | 1.4k | ## Licenses | License | Datasets | Commercial use | | --- | --- | --- | | cc-by-4.0 | 14 | Yes | | other | 6 | Read the license | | not stated | 5 | Not stated | | cc-by-nc-4.0 | 5 | Not without separate permission | | cc-by-3.0 | 2 | Read the license | | cdla-sharing-1.0 | 1 | Read the license | | cc0-1.0 | 1 | Read the license | | apache-2.0 | 1 | Yes | ## Who Publishes Them | Publisher | Datasets | | --- | --- | | [Shrikant Nayak](https://savrn.com/model-publishers/shrikanth-19) | 6 | | [The Larry Project](https://savrn.com/model-publishers/thelarryproject) | 2 | | [Thant Zin Phyo](https://savrn.com/model-publishers/thantzinphyo) | 2 | | [Ahmed Wasfy](https://savrn.com/model-publishers/oddadmix) | 2 | | [Kapture CX](https://savrn.com/model-publishers/kapturecx) | 2 | | [Muhammad Zohaib Hassan](https://savrn.com/model-publishers/zaibihassan) | 1 | ## All 35 Datasets Dataset · Speech recognition ### [fleurs](https://savrn.com/datasets/fleurs) [Google](https://savrn.com/model-publishers/google) Universal Representations of Speech](https://arxiv.org/abs/2205.12446) Fleurs is the speech version of the FLoRes machine translation benchmark. We use 2009 n-way parallel sentences from the FLoRes dev and devtest publicly available sets, in 102 languages. Training sets have around 10 hours of supervision. Speakers of the train sets are different than speakers from the dev/test sets. Multilingual fine-tuning is used and ”unit error rate” (characters, signs) of all languages is averaged. Languages and results are also grouped into seven geographical areas: The datasets library allows you to load and pre-process your dataset in pure Python, at scale. The dataset can be downloaded and prepared… Publicly accessible cc-by-4.0 10K= 0.3 s, so cutting on caption boundaries puts most cuts inside continuous speech. Snapping caption boundaries to the quietest nearby window does not… Publicly accessible cc-by-3.0 [View dataset](https://savrn.com/datasets/unnamed-split-ar) Dataset · Speech recognition ### [dialectal-arabic-voices](https://savrn.com/datasets/dialectal-arabic-voices) [Moaead](https://savrn.com/model-publishers/moaead) An expanding collection of Arabic audio from YouTube, SoundCloud, and other sources. Currently labelled Palestinian Arabic (ps). 53,194 recordings · approximately 9,573.3 hours · 501.14 GB Audio retains its original format and sample rate. Recordings have not yet been segmented or individually checked for dialect. New audio is added in ~5 GB Parquet files. The final incomplete file is held locally until it fills. Publicly accessible [View dataset](https://savrn.com/datasets/dialectal-arabic-voices) Dataset · Speech recognition ### [ralomi-reazonspeech-20h-development](https://savrn.com/datasets/ralomi-reazonspeech-20h-development) [TTS Dataset](https://savrn.com/model-publishers/tts-dataset) This repository is configured as public with manual access approval. No audio, transcripts, or training manifests have been uploaded yet. The original transfer-staging metadata is retained below as history. Since 2026-10-08 this repository also preserves encrypted saved fullstate checkpoints from the actual 5000h development training and its continuations. Source: ReazonSpeech, provided by Reazon Human Interaction Laboratory / Reazon Holdings, Inc. Upstream specifies use solely within Article 30-4 of the Japanese Copyright Act; review the upstream dataset card and comply with applicable law. The license is not altered by this access gate. No additional license restrictions are asserted… Access requested at publisher cdla-sharing-1.0 [View dataset](https://savrn.com/datasets/ralomi-reazonspeech-20h-development) Dataset · Speech recognition ### [DNU-ViSpeech](https://savrn.com/datasets/dnu-vispeech) [Trang Lê](https://savrn.com/model-publishers/trangle2108) DNU-ViSpeech is a Vietnamese scripted/read-speech corpus for automatic speech recognition (ASR). This release contains 31,070 utterances from 26 anonymized speakers, totaling 75.570 hours. Audio is WAV, 16 kHz, mono, uncompressed PCM 16-bit. The primary protocol is DNU-ASR-v1.1. Its evaluation split combines unseen-speaker and recording-configuration shift; those two factors cannot be isolated with the current data. The corpus supports research on Vietnamese ASR adaptation, especially whether compact models can approach larger zero-shot systems after domain fine-tuning on read speech. The repository follows the Hugging Face AudioFolder layout. Each split has its own metadata.csv, and… Publicly accessible cc-by-nc-4.0 10K