Samromur Unverified 22.07

  Authors

Hedström, Staffan ; et al.show everyone

Hedström, Staffan ; Fong, Judy Y. ; Þórhallsdóttir, Ragnheiður ; Mollberg, David Erik ; Guðmundsson, Smári Freyr ; Jónsson, Ólafur Helgi ; Þorsteinsdóttir, Sunneva ; Magnúsdóttir, Eydís Huld ; Gudnason, Jon

  Item identifier

http://hdl.handle.net/20.500.12537/265

 Project URL

https://github.com/cadia-lvl/samromur

 Demo URL

http://openslr.org/128/

 Date issued

2022-07-01

 Type

audio, corpus

 Size

2233 hours, 2159314 utterances, 125 gb

 Language(s)

Icelandic

 Description

This release of data from the Samrómur collection contains all the collected data not present in other releases. The data is mostly UNVERIFIED. It contains 2,159,314 (2233 hours) speech-recordings in Icelandic, of which 84,161 have been verified. Ca 700,000 utterances have been scored with marosijo, this score indicates how likely the audio is to match the transcript. For more information about marosijo, please see [1] and [2]. The corpus contains 17,984 unique speakers. The corpus is NOT split into train, dev and test subsets. For such subsets please look at other Samrómur releases. All demographics are self reported. The dataset contains folders that correspond to speaker IDs, and the audio files inside use the following naming convention: {speaker_ID}-{utterance_ID}.flac. Average utterance length is 3.7 seconds. Þessi útgáfa af gögnum úr safni Samróma eru öll söfnuð gögn sem ekki eru til í öðrum útgáfum. Gögnin eru að mestu óyfirfarin. Útgáfan inniheldur 2.159.314 (2.233 klst.) talupptökur á íslensku, þar af 84.161 sem hafa verið staðfest. Um 700.000 hafa verið skoruð með marosijo sem gefur til kynna hvort líklegt sé að það sé gilt eða ekki. Fyrir frekari upplýsingar um Marosijo sjá [1] og [2]. Málheildin inniheldur 17.984 mismunandi raddir og hefur EKKI verið skipt upp í þjálfunar- (train), þróunar- (dev) og prófunarsett (test). Allar lýðfræðilegar upplýsingar hafa notendur sjálfir slegið inn. Hvert sett inniheldur möppur sem samsvara auðkenni raddar. Hljóðskrárnar nota eftirfarandi nafnareglur: {speaker_ID}-{utterance_ID}.flac. Meðallengd upptöku er 3,7 sekúndur. [1] Gudason et al., "Building ASR corpora using Eyra" https://www.isca-speech.org/archive/pdfs/interspeech_2017/gunason17_interspeech.pdf [2] Guðmundsson, Smári Freyr, "Samrómur automated verification wrapup" https://github.com/cadia-lvl/samromur-tools/tree/master/QualityCheck

 Publisher

Reykjavik University

 Acknowledgement

Ministry of Education, Science and Culture

Project code: Data recording using Eyra/Samrómur (H1)

Project name: Language Technology for Icelandic 2019-2022

 Subject(s)

audio corpus automatic speech recognition speaker verification speaker identification

 Collection(s)

Clarin IS

Sýna fulla færslu atriðis