Loading...
Loading...
Speech recognition, text-to-speech, speaker identification, music classification, audio event detection
2,585 datasets
A 2.4 GB multimodal dataset integrates EEG brain activity data with expressive movement and music audio. It was created with support from the National Science Foundation, NIH, and the University of Texas MD Anderson Cancer Center. The dataset is structured for research on the coupling of neurological signals, physical motion, and auditory stimuli.
1,000 hours of Arabic speech audio sampled at 16 kHz, crawled from over 700 YouTube channels. The MASC dataset is multi-regional, multi-genre, and multi-dialect, created by MohamedRashad and last updated in April 2026. It is intended to advance research and development in Arabic speech technology.
Geospatial polygon features represent isopach and lithofacies data for Carboniferous strata in the Western Canada Sedimentary Basin. The shapefiles were created by the Alberta Geological Survey from mid-1990s digital files and edited in 2005-2006. This dataset is part of the Geological Atlas of the Western Canada Sedimentary Basin.
Stoddart/Mattson/Kibbey Isopach and Lithofacies GIS data contains line features representing subsurface geological formations from the Carboniferous period in the Western Canada Sedimentary Basin. The dataset was created by the Alberta Geological Survey from archived digital files produced in the mid-1990s and later edited in 2005-06. It is part of the Geological Atlas of the Western Canada Sedimentary Basin, specifically Chapter 14, Figure 35.
Ng'akarimojong (kdj), an Eastern Nilotic language with approximately 370,000 speakers in Karamoja, Uganda, is the focus of this speech dataset. It was created by Speedykom using GRN recordings segmented via silence detection. Audio files are in WAV format at 16 kHz mono, paired with UTF-8 transcripts auto-generated by the facebook/mms-1b-all model with a Teso adapter.
BASS is a benchmark dataset for evaluating music understanding and reasoning in audio language models. It comprises 2,658 questions across 12 tasks and 4 categories, covering 1,993 unique songs and over 138 hours of music. The dataset was created by author 'oreva' and last updated on 2026-04-08.
A case study dataset for text extraction and NLP analysis. The dataset likely contains textual data for analysis tasks. Its origin, size, and specific contents are not detailed in the provided metadata.
A dataset for text-to-speech (TTS) in the Nepali language, likely containing audio samples and corresponding text transcriptions. It was published on the HuggingFace platform by an author named Titung. The dataset listing was last updated on June 4, 2026.
A dataset titled 'urdu-tts-processed' is hosted on Kaggle. The dataset likely contains processed audio and text data for Urdu language speech synthesis. Metadata is minimal; the specific content, scale, and creation details require verification after download.
A dataset for Automatic Speech Recognition (ASR) tasks, published on Kaggle. The dataset's specific content, size, and origin are not detailed in the available metadata. Its intended use is likely for training or evaluating custom speech recognition models.
Music Extract Wav2Vec2 Features is a dataset published on Kaggle. The title suggests it contains extracted features from audio files, likely using the Wav2Vec2 model. Metadata is minimal; actual content requires verification after download.
Voice Annotation Data v2 is a curated dataset of 18,632 audio samples, comprising 9,391 positive and 9,241 negative examples across 58 voice dimensions. The dataset was created by TTS-AGI and was last updated on April 15, 2026. Each dimension includes up to 25 positive examples of audio that fits the category and 25 negative examples confirmed not to fit, using the Gemini 2.0 Flash model.
Music Arena provides an open platform dataset for evaluating text-to-music models. The dataset is hosted on Hugging Face by the music-arena organization and was last updated in April 2026. It likely contains audio files and associated metadata for benchmarking AI music generation systems.
A parsed and formatted voice dataset containing recordings and text transcripts in Standard Moroccan Amazigh. The dataset is intended for training Automatic Speech Recognition and Text-to-Speech models and was published by the Tamazight-NLP organization. The dataset page was last updated on April 21, 2026.
A speech recognition dataset associated with the Google WAXAL ASR Challenge. The title suggests it contains audio data for the Luganda language. It is hosted on the Kaggle platform, but detailed metadata about its size, format, and creation is unavailable.
A Kaggle challenge dataset likely containing audio data for speech recognition tasks. The dataset's specific content, size, and origin are not detailed in the available metadata. Further details such as the number of samples, recording conditions, and annotation specifics require verification after download.
LibriSpeech-Mixtures is an audio dataset hosted on Kaggle, likely derived from the LibriSpeech corpus. The dataset appears to contain mixtures of speech signals, which are commonly used for tasks like source separation. Specific details on the number of files, duration, and creation methodology are not provided in the available metadata.
XRXRX aggregated this multilingual speech dataset from seven distinct sources, including Multilingual LibriSpeech, VoxPopuli, and GigaSpeech 2. The collection was last updated on April 12, 2026. Each constituent dataset retains its own license, with most permitting commercial use.
SilentWear provides surface electromyography (EMG) data recorded from a wearable neckband for both vocalized and silent speech. The dataset, created by PulpBio and last updated in April 2026, is designed to support research in ultra-low-power wearable AI systems. It likely contains multi-session recordings intended for decoding speech from muscle signals.
75 hours of Filipino Tagalog speech audio across 639 files, provided in MP3 and WAV formats. The dataset was created by Speech-data and was last updated in March 2026. It contains recordings from a speaker pool that is 55% male and 45% female, with ages ranging from 18 to over 50 years.