Speech and Computational Pragmatics and Semantics for AI

If you can donate audio samples to our project, we need:

  • WAV PCM with at least 44 kHz/16 Bit recorded using your mobile phone, laptop, tablet, or recording device
  • Ensure the recording is clear with minimal to no background noise (Recommended to use a mobile phone with microphone held at some distance from mouth).

File naming convention

The file name is constructed:

USERID_DATE_UTTERANCEID_FROM_TO.wav
  • USERID (2xCHAR) is a two character ID.
  • DATE is in the format YYYYMMDD
  • UTTERANCEID, unique id of each utterance as per this google doc
  • FROM (int) is the index in Python notation
  • TO (int) is the index in Python notation (including the token before the int)

We will focus here on contrastive stress on:

  • subject
  • verb
  • object
  • locative preposition phrase (or other such predicative modifiers)

Each recording consists of a plain declarative intonation using the unmarked or default word order, an answer to a question like "what happened":

  • I ride a bike to class.

    FileName:AA_20180711_5_0_0.wav

Then we generate all contrastive alternations:

  • I say that "he" (rides a bike to class), and you answer: No, "I" (ride a bike to class)
  • I say "fly" (I __ a bike to class), and you answer: No, I RIDE a bike to class.

So all variants will be:

  • I ride a bike to class.

    FileName:AA_20180711_5_0_1.wav
  • I ride a bike to class.

    FileName:AA_20180711_5_1_2.wav
  • I ride a bike to class.

    FileName:AA_20180711_5_3_4.wav
  • I ride a bike to class.

    FileName:AA_20180711_5_5_6.wav
Please create a folder with your userid in the following box shared location, and upload your recordings there
Prosody-Pragmatics/Corpus[This IU Box, in case you face any issues, please contact us.]
E.g.: Arnav has userid AA, so you will find all his recordings uploaded to a folder "AA"

In the above example, the files belong to user Arnav, who has chosen userid AA, these files were completed on 11 July 2018, the "5" is the utterance id and the integer range after that is the stressed word range.

Please add new uttrances to the google doc.