Hugging Face's logo

Datasets:
meetween
/
mumospee

Tasks:
Text-to-Speech
Automatic Speech Recognition
Modalities:
Text
Formats:
parquet
Languages:
English
Chinese
Japanese
Size:
10M - 100M
Tags:
Speech
Video
Libraries:
Datasets
Dask
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Mumospee: A MUltiMOdal SPEEch Corpus

The Mumospee dataset supports the Meetween project's mission of enabling inclusive, language-neutral collaboration across virtual environments. The release provides metadata and download URLs for a curated collection of speech audio sourced from publicly available datasets, optimized for processing on high-performance computing clusters.

Mumospee Overview

Mumospee is a comprehensive multilingual speech-metadata corpus featuring:

  • 224,117 hours of speech metadata across 65,757,486 samples
  • Coverage of 25 EU languages plus a long tail of additional languages
  • Collections drawn from existing datasets in different speaking styles and content genres:
_TAGS = ["CoVoST", "GigaSpeech", "PeopleSpeech", "Librispeech", "LibriTTS", "Emilia", "MOSEL"]

A smaller version with fewer than 1000 rows is also available as mumospee_small for testing purposes.

Dataset Statistics

Overview (All Splits Combined)

Metric Value
Total samples 65,757,486
Total audio duration 224,117h 20m 23.0s (224,117.3h)
Average duration per sample 12.27s
Avg transcript length 21.0 words
Total parquet shards 35

Per-Split Overview

Split # Samples Duration Avg Duration Avg Words Shards
train 64,995,066 222,881h 24m 09.6s (222,881.4h) 12.35s 21.1 33
test 398,191 639h 06m 04.1s (639.1h) 5.78s 10.5 1
validation 364,229 596h 50m 09.3s (596.8h) 5.90s 10.4 1

Language Distribution

Value train samples train % test samples test % validation samples validation % Total samples Total % Total Duration Total Dur %
en 30,354,591 46.70% 302,179 75.89% 269,449 73.98% 30,926,219 47.03% 72,634h 44m 09.5s 32.41%
zh 19,969,304 30.72% 0 0.00% 0 0.00% 19,969,304 30.37% 49,922h 33m 08.9s 22.28%
de 1,288,162 1.98% 13,511 3.39% 13,511 3.71% 1,315,184 2.00% 6,264h 10m 16.0s 2.80%
fr 1,241,292 1.91% 14,760 3.71% 14,760 4.05% 1,270,812 1.93% 6,129h 37m 57.4s 2.74%
ja 870,783 1.34% 684 0.17% 635 0.17% 872,102 1.33% 1,718h 30m 51.3s 0.77%
es 629,900 0.97% 13,221 3.32% 13,221 3.63% 656,342 1.00% 4,639h 34m 39.7s 2.07%
it 588,543 0.91% 8,183 2.06% 8,940 2.45% 605,666 0.92% 4,614h 59m 36.9s 2.06%
pt 571,574 0.88% 4,023 1.01% 3,318 0.91% 578,915 0.88% 4,452h 55m 29.7s 1.99%
nl 571,403 0.88% 1,699 0.43% 1,699 0.47% 574,801 0.87% 4,501h 00m 46.2s 2.01%
sv 567,993 0.87% 0 0.00% 0 0.00% 567,993 0.86% 4,511h 52m 00.0s 2.01%
pl 567,309 0.87% 0 0.00% 0 0.00% 567,309 0.86% 4,517h 07m 24.8s 2.02%
lv 563,787 0.87% 1,629 0.41% 1,125 0.31% 566,541 0.86% 4,440h 37m 21.7s 1.98%
cs 565,495 0.87% 0 0.00% 0 0.00% 565,495 0.86% 4,517h 07m 22.8s 2.02%
ro 563,319 0.87% 0 0.00% 0 0.00% 563,319 0.86% 4,481h 04m 42.5s 2.00%
fi 563,305 0.87% 0 0.00% 0 0.00% 563,305 0.86% 4,472h 22m 31.4s 2.00%
sl 560,950 0.86% 360 0.09% 509 0.14% 561,819 0.85% 4,389h 53m 19.1s 1.96%
et 558,644 0.86% 1,571 0.39% 1,576 0.43% 561,791 0.85% 4,362h 52m 02.6s 1.95%
hu 559,069 0.86% 0 0.00% 0 0.00% 559,069 0.85% 4,390h 00m 33.0s 1.96%
mt 556,919 0.86% 0 0.00% 0 0.00% 556,919 0.85% 4,361h 28m 18.0s 1.95%
el 556,641 0.86% 0 0.00% 0 0.00% 556,641 0.85% 4,398h 06m 42.3s 1.96%
da 553,516 0.85% 0 0.00% 0 0.00% 553,516 0.84% 4,322h 44m 48.0s 1.93%
bg 553,022 0.85% 0 0.00% 0 0.00% 553,022 0.84% 4,325h 19m 31.7s 1.93%
lt 551,298 0.85% 0 0.00% 0 0.00% 551,298 0.84% 4,297h 54m 19.8s 1.92%
sk 545,302 0.84% 0 0.00% 0 0.00% 545,302 0.83% 4,315h 11m 03.5s 1.93%
hr 339,551 0.52% 0 0.00% 0 0.00% 339,551 0.52% 2,683h 37m 11.3s 1.20%
ko 92,182 0.14% 0 0.00% 0 0.00% 92,182 0.14% 217h 09m 58.0s 0.10%
ca 54,173 0.08% 12,730 3.20% 12,730 3.50% 79,633 0.12% 119h 48m 09.3s 0.05%
ru 12,112 0.02% 6,300 1.58% 6,110 1.68% 24,522 0.04% 38h 40m 15.3s 0.02%
zh-CN 7,085 0.01% 4,898 1.23% 4,843 1.33% 16,826 0.03% 26h 37m 10.2s 0.01%
fa 4,347 0.01% 3,445 0.87% 3,445 0.95% 11,237 0.02% 14h 20m 32.8s 0.01%
tr 3,966 0.01% 1,629 0.41% 1,624 0.45% 7,219 0.01% 7h 52m 17.3s 0.00%
mn 2,018 0.00% 1,759 0.44% 1,761 0.48% 5,538 0.01% 8h 21m 37.1s 0.00%
ar 2,029 0.00% 1,695 0.43% 1,758 0.48% 5,482 0.01% 5h 35m 01.6s 0.00%
sv-SE 2,160 0.00% 1,595 0.40% 1,349 0.37% 5,104 0.01% 4h 24m 34.4s 0.00%
id 1,243 0.00% 844 0.21% 792 0.22% 2,879 0.00% 2h 58m 58.9s 0.00%
ta 1,358 0.00% 786 0.20% 384 0.11% 2,528 0.00% 3h 04m 21.3s 0.00%
cy 721 0.00% 690 0.17% 690 0.19% 2,101 0.00% 3h 01m 18.5s 0.00%

Tag / Source Distribution

Value train samples train % test samples test % validation samples validation % Total samples Total % Total Duration Total Dur %
Emilia 40,237,834 61.91% 0 0.00% 0 0.00% 40,237,834 61.19% 101,585h 04m 02.8s 45.33%
MOSEL 12,679,032 19.51% 0 0.00% 0 0.00% 12,679,032 19.28% 100,316h 58m 51.7s 44.76%
GigaSpeech 6,016,616 9.26% 19,931 5.01% 5,715 1.57% 6,042,262 9.19% 7,544h 42m 14.1s 3.37%
CoVoST 3,925,255 6.04% 327,848 82.33% 323,985 88.95% 4,577,088 6.96% 7,114h 56m 33.1s 3.17%
PeopleSpeech 1,501,271 2.31% 34,898 8.76% 18,622 5.11% 1,554,791 2.36% 5,987h 42m 22.4s 2.67%
LibriTTS 353,817 0.54% 9,955 2.50% 10,340 2.84% 374,112 0.57% 585h 37m 48.6s 0.26%
Librispeech 281,241 0.43% 5,559 1.40% 5,567 1.53% 292,367 0.44% 982h 18m 30.3s 0.44%

License Distribution

Value train samples train % test samples test % validation samples validation % Total samples Total %
CC-BY-NC-4.0 40,237,834 61.91% 0 0.00% 0 0.00% 40,237,834 61.19%
CC-BY-4.0 13,314,090 20.48% 15,514 3.90% 15,907 4.37% 13,345,511 20.30%
apache-2.0 6,016,616 9.26% 19,931 5.01% 5,715 1.57% 6,042,262 9.19%
CC0 3,925,255 6.04% 327,848 82.33% 323,985 88.95% 4,577,088 6.96%
CC-BY;CC-BY-SA 1,501,271 2.31% 34,898 8.76% 18,622 5.11% 1,554,791 2.36%

apache-2.0 rows are the GigaSpeech subset: its transcripts/manifest are Apache-2.0, but the audio is governed by the GigaSpeech Data User Agreement — non-commercial research only, no raw-audio redistribution.

Data quality notes

  • MOSEL audio is hosted separately at meetween/mumospee_mosel (VoxPopuli PLENARY sessions, 23 EU languages); each row's url points to the whole-session shard there.
  • Language labels are those declared by each source and are not independently verified; for MOSEL they come from VoxPopuli session-level metadata (~1.5% disagree with a fastText check).
  • 252 CoVoST rows have duration = NaN (unreadable source audio) — filter with df["duration"].notna(); all other rows have a valid numeric duration.

Mumospee dataset structure

Each row in the metadata represents one audio sample with the following fields:

  • path: the relative path of the audio file
  • url: the link to download the parquet shard containing the audio
  • type: the sample type (audio or video)
  • duration: duration in seconds
  • language: language of the audio
  • transcript: transcript text
  • tag: origin dataset (one of _TAGS above)
  • split: train, test, or validation
  • license: license governing this sample

Example row:

{
  "path": "3660-172183-0000.flac",
  "url": "https://huggingface.co/datasets/meetween/mumospee_librispeech/resolve/main/librispeech-parquet/dev-other.parquet",
  "type": "audio",
  "duration": 5.405,
  "language": "en",
  "transcript": "GERAINT AS HE HAD BEEN USED TO DO WHEN HE WAS AT ARTHUR'S COURT FREQUENTED TOURNAMENTS",
  "tag": "Librispeech",
  "split": "validation",
  "license": "CC-BY-4.0"
}

Intended Uses

This dataset is designed to enable SpeechLLM and other large language models to support language-neutral virtual meeting applications.

Data Sources

The release includes metadata and download URLs for the following publicly available datasets:

Example usage 

# pip install datasets

from datasets import load_dataset

# ── Load all splits at once ───────────────────────────────────────────────────

dataset = load_dataset("meetween/mumospee")
print(dataset)
# DatasetDict({
#     train:      Dataset({features: [...], num_rows: ...})
#     test:       Dataset({features: [...], num_rows: ...})
#     validation: Dataset({features: [...], num_rows: ...})
# })

# ── Load a specific split ─────────────────────────────────────────────────────

train_data      = load_dataset("meetween/mumospee", split="train")
test_data       = load_dataset("meetween/mumospee", split="test")
validation_data = load_dataset("meetween/mumospee", split="validation")

License

The metadata is published under CC-BY-4.0. Each individual sample is governed by its own license, recorded per-row in the license column. Users must comply with the licensing terms of each underlying dataset.

Changelog

2026-07-02

  • Repackaged the corpus into sharded Parquet for faster, streaming-friendly loading.
  • Extended CoVoST to the full CoVoST 2 catalogue.
  • Refreshed the GigaSpeech, LibriTTS, and MOSEL subsets from their source datasets — correcting transcripts, licenses, and audio download links, and expanding coverage.
  • Recomputed all dataset statistics from the released data.
Downloads last month
665
Total file size:
4.84 GB