ENRICH: EU Project 675324 Marie Curie (MSCA) Innovative Training Network
2016-2020

Greek Harvard Sentences

A Harvard-based corpus for speech technology and audiology

The current material consists of 720 sentences of variable syntactic structure, designed according to the following criteria. Each sentence comprises exactly 5 keywords which are almost always content words, and (optionally) 1 to 4 non-keywords which are pronouns and other function words; hence total sentence length ranges strictly from 5 to 9 words. All words contain one, two or -maximally- three syllables, and have been selected so that the sentences are meaningful and resemble everyday language. Keywords have been combined so that the sentences are semi-predictable. Although a number of the original Harvard sentences have been translated into Greek, the majority of the sentences in the present corpus are authentic.

This is still work in progress, so please check for updates and additional recordings.

You can find below:

• A WORD file that contains the 720 sentences as they should appear to listeners.


• An EXCEL file that contains the Greek orthography, the phonetic transcription and important meta-data such as number of words, syllables and phonemes per sentence as well as which are the keywords and how many syllables they contain.


• As the material is not balanced yet, this information may be useful in deciding which sentences to select for intelligibility tests.


• A ZIP file that contains recordings of the material: one female speaker and one male speaker (raw data in .wav format).

To cite this work:

Sfakianaki, A. "Designing a Modern Greek sentence corpus for audiological and speech technology research". TO APPEAR: In Proc. of the 14th International Conference on Greek Linguistics (ICGL14), September 5-8, 2019, University of Patras, Greece. [Unpublished PDF]

Conference website: https://icgl14.events.upatras.gr/

Material:

• WORD file with GrHarvard sentences: [WORD File]


• EXCEL file - orthography, transcription, meta-data: [EXCEL File]


• Abstract and PPT file of the conference presentation here.


• Full paper to appear in forthcoming proceedings (for unpublished manuscript, please see this PDF).
  
  
• Link (raw data in .WAV format): [DATASET] - v.03 (1/10/2020)