14

I am working on building a language classifier in speech/audio samples. I have been trying to find a dataset which may have considerable number of speech samples in various languages. The audio files maybe of any standard format like wav, mp3 etc. containing human voice/conversation with least amount of background noise/music.

I am unable to find any such dataset. Can someone share link of any speech dataset that may be good for this research.

9

You can use the Tatoeba website which has full sentences in text and audio as downloads.

Sentences with audio

Download

http://downloads.tatoeba.org/exports/sentences_with_audio.tar.bz2

Fields and structure

Sentence id 

File description

Contains the ids of the sentences, in all languages, for which audio is available. 

Thanks to Nicolas Raoul in this answer.

7

I suggest looking at the SpokenLanguages2 dataset, which is very extensive, containing a bit more than one hour of speech for each of the 172 languages: https://community.topcoder.com/longcontest/?module=ViewProblemStatement&rd=16555&pm=13978

The audio is very high quality. Although they do not provide access to the testing set, you have to do your own split.

6

UCLA Phonetics Lab Archive offers many audio files of spoken word in many languages, although you'll have to dig through and mix-match to your liking:
http://archive.phonetics.ucla.edu/
IPA Handbook downloads actually seems to be exactly what you are asking for, however its only for personal use; any other usage requires contacting them for permission...so not 100% open data:
https://www.internationalphoneticassociation.org/content/ipa-handbook-downloads

6

Maybe you can do something with https://rhinospike.com/language/ It's a bit like Tatoeba mentioned above

edit : Take also a look at librivox (public domain audiobooks). It comes with an API, then you could send a request like https://librivox.org/api/feed/audiobooks?fields={url_zip_file,language}&offset=XXX&limit=YYY to extract files' id

  • Feel free to post an answer for each. Also, please specify about how many languages each one covers. Thanks! :-) – Nicolas Raoul Mar 1 '17 at 7:47
5

I've found something else. It is called the wide language index. See on github. It consists in a listing of radio podcasts. It's pretty impressive in the numbers of languages covered.

Note : It was made for the game The Great language game which is fun ;).

1

Maybe you can check this website, it collects many audio-related datasets.

https://www.audiocontentanalysis.org/data-sets/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.