{"success":true,"data":{"author":"ESPnet / CMU WAVLab","bucketAccessType":"public","bucketName":"espnet/floras","bucketPrefix":"","category":"Speech & Audio","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:17.759Z","description":"FLORAS is a 50-language benchmark For LOng-form Recognition And Summarization of spoken language. Unlike pre-segmented, single-speaker read-speech benchmarks such as LibriSpeech and FLEURS, it tests models on raw long-form conversational audio with one or many speakers. It provides one- to three-way parallel data for long-form automatic speech recognition, X-to-English speech translation and speech summarization, so some samples have only speech and transcripts while others add translations and summaries; in total it contains roughly 32,000 hours of raw audio. FLORAS is derived from YODAS, the Creative Commons YouTube crawl, filtered with CTC alignment scores to remove noisy data; translations and summaries were produced by pseudo-labelling with a large language model, as the card describes. This repository holds two configurations, monolingual and multilingual, as Parquet shards (922 files in monolingual; the multilingual folder lists at least 1,000) with 16 kHz audio and text fields (measured 2026-11-14; the Hub reports about 3 TB). Load with datasets.load_dataset('espnet/floras', 'monolingual', streaming=True). It inherits the CC BY 3.0 licence of its YODAS source, as stated in the YODAS paper and on the card.","displayTitle":"FLORAS: long-form speech recognition, translation and summarization benchmark","id":"3353ba0a-080f-5afe-bce7-26c2700c15e4","license":"CC-BY-3.0","licenseEvidenceUrl":"https://arxiv.org/pdf/2406.00899","objectCountEstimate":7138,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"espnet/floras\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"espnet/floras\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=espnet, license:cc-by-3.0, size_categories:10K<n<100K, format:parquet, downloads=4416"],"format":"mixed","generatedAt":"2026-11-14T14:15:20.233Z","layout":{"example":"monolingual/dev-00000-of-00008.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"monolingual/dev-00000-of-00008.parquet","truncatedAt":512}],"sampleObjects":["monolingual/dev-00000-of-00008.parquet","monolingual/dev-00001-of-00008.parquet","monolingual/dev-00002-of-00008.parquet","monolingual/dev-00003-of-00008.parquet","monolingual/dev-00004-of-00008.parquet"],"sidecars":[{"bytes":"---\nlicense: cc-by-3.0\ndataset_info:\n- config_name: monolingual\n  features:\n  - name: id\n    dtype: string\n  - name: language\n    dtype: string\n  - name: score\n    dtype: string\n  - name: audio\n    dtype:\n      audio:\n        sampling_rate: 16000\n  - name: text\n    dtype: string\n  - name: summary\n    dtype: string\n  - name: translation\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 2250087924\n    num_examples: 50814\n  - name: dev\n    num_bytes: 3730403898.0\n    num_examples: 81\n  - name: test\n    num_bytes: 6882657690.0\n    num_examples: 116\n  download_size: 27806858743\n  dataset_size: 21226123202.0\n- config_name: multilingual\n  features:\n  - name: id\n    dtype: string\n  - name: language\n    dtype: string\n  - name: score\n    dtype: string\n  - name: audio\n    dtype:\n      audio:\n        sampling_rate: 16000\n  - name: text\n    dtype: string\n  - name: summary\n    dtype: string\n  - name: translation\n    dtype: string\n  splits:\n  - name: dev\n    num_bytes: 49979924635.32\n    num_examples: 1154\n  - name: test\n    num_bytes: 56817933774.28\n    num_examples: 1188\n  download_size: 102717641464\n  dataset_size: 106797858409.6\nconfigs:\n- config_name: monolingual\n  data_files:\n  - split: train\n    path: monolingual/train-*\n  - split: dev\n    path: monolingual/dev-*\n  - split: test\n    path: monolingual/test-*\n- config_name: multilingual\n  data_files:\n  - split: dev\n    path: multilingual/dev-*\n  - split: test\n    path: multilingual/test-*\ntask_categories:\n- automatic-speech-recognition\n- translation\n- summarization\nlanguage:\n- en\n- es\n- fr\n- de\n- nl\n- it\n- pt\n- hu\n- fi\n- el\n- ca\n- eo\n- et\n- da\n- la\n- sv\n- cy\n- gl\n- ru\n- pl\n- uk\n- ro\n- cs\n- sl\n- sk\n- hr\n- bg\n- bs\n- ka\n- tr\n- fa\n- ar\n- uz\n- az\n- ku\n- ky\n- hi\n- ta\n- ur\n- bn\n- id\n- vi\n- th\n- mi\n- ms\n- ja\n- zh\n---\n\n# FLORAS\n\nFLORAS is a 50-language benchmark **F**or **LO**ng-form **R**ecognition **A**nd **S**ummarization of spoken language. \nThe goal of FLORAS is to create a more realistic benchmarking environment for speech recognition, translation, and summarization models. \nUnlike typical academic benchmarks like LibriSpeech and FLEURS that uses pre-segmented single-speaker read-speech, FLORAS tests the capabilities of models on raw long-form conversational audio, which can have one or many speakers.\n\nTo encourage research in multi-tasking, FLORAS provides 1-way to 3-way parallel data for long-form Automatic Speech Recognition (ASR), long-form X-to-EN Speech Translation (ST), and Speech Summarization (SSUM).\nThis means that some samples only have paired speech and transcripts, while others may have paired speech, transcripts, translations and/or summaries.\nIn total, FLORAS contains roughly 32,000 hours of raw audio.\n\n## Dataset Creation\n\nFLORAS is derived from [YODAS](https://huggingface.co/datasets/espnet/yodas), a large multilingual crawl of YouTube videos and their subtitles. \nSince the raw crawl of YODAS is too noisy for direct training in many settings, we filter out most of the data using CTC alignment scores.\nThe translations and summaries are obtained via pseudo-labelling using Google's [Gemini Flash](https://deepmind.google/technologies/gemini/flash/).\nOur translators then filtered out or corrected faulty pseudo-labels in the test set. We did not perform filtering on the training/development sets.\n\n## Dataset Structure\n\nFLORAS is organized into two subsets, each with data splits for training, validation, and testing.\n```\nFLORAS\n- monolingual\n  - train\n  - dev\n  - test\n- multilingual\n  - train\n  - dev\n  - test_unverified\n  - test_verified\n```\nThe monolingual subset contains English-only data. The multilingual subset contains the data for the other 49 languages.\n\nThe multilingual subset contains two test sets: `test_unverified` and `test_verified`. \n\nVerified languages are those that have had professional translators and/or native speakers verify the translation/summary pseudo-labels.\n\nUnverified languages are those that did not go through this process (See below to determine which languages have been verified).\n\n## Data Fields\n\nEach subset/split has the following da","encoding":"utf8","key":"README.md","name":"README.md","size":5714,"truncated":true}],"sourceObjectKey":"monolingual/dev-00000-of-00008.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"espnet/floras\")\nprint(ds)","starterSamplePath":"monolingual/dev-00000-of-00008.parquet","totalObjects":7138},"summary":"FLORAS is a 50-language benchmark For LOng-form Recognition And Summarization of spoken language.","title":"FLORAS: long-form speech recognition, translation and summarization benchmark","updatedAt":"2026-11-14T17:49:03.488Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/espnet/floras"}}