About the Dataset

308 hours

This audio dataset contains 308 hours of Arabic Speech Data in Banking, Telecommunication, Insurance and Retail domains, recorded by speakers from Egypt, Jordan, UAE, and Yemen.

There are 102 hours of Arabic (Egypt) human-to-human audio with the following domain distribution per dataset:

20.85 hours of Banking
30.65 hours of Insurance
24.32 hours of Retail
26.88 hours of Telecommunication

There are 90 hours of Arabic (Jordan) human-to-human audio with the following domain distribution per dataset:

24.15 hours of Banking
23.88 hours of Insurance
21.32 hours of Retail
21.5 hours of Telecommunication

There are 25 hours of Arabic from the United Arab Emirates human-to-human audio with the following domain distribution per dataset:

13.05 hours of Retail
12.68 hours of Telecommunication

There are 91 hours of Arabic (Yemen) human-to-human audio with the following domain distribution per dataset:

33.55 hours of banking human-to-human audio recordings
31 hours of Insurance human-to-human audio recordings
11.25 hours of retail human-to-human audio recordings
15.53 hours of telecommunication human-to-human audio recordings

Defined.ai creates scenarios for our crowd members to follow, which they study beforehand. They then record a conversation, one speaker playing the agent, the other speaker “playing out” the scenario with spontaneous content. The recording is done via telephony and is saved in 8kHz 16 bit per channel. That content is then transcribed.

The dataset is covered by Defined.ai's standard license agreement. The license agreement is perpetual and allows for the commercialization of all models built on the data.

Other characteristics: