Voice recognition: teaching AI to speak and listen

Company is a leading micro tasking marketplace, catering data management and web research services as well as AI algorithms training.

Countries:

Netherlands, Portugal, Russia

Tests performed:

2,600

Client’s field:

AI development

Research period:

Jan 2019 – July 2019

Client:

NDA

Cooperation duration:

2 years

Number of testers:

1, 620

The challenge

The client was developing the new software in the field of artificial intelligence, which aim was to provide the customers with voice recognition option.

However, there was no possibility to calibrate it for Russian, Dutch and Portuguese languages.

Client’s first decision was to develop the app to crowd source the needed voice samples for AI further learning, which didn’t bring the desired results due to lack of systematical approach and people’s segmentation.

Their second decision was to contact us.

The current challenges of speech recognition are caused by two major factors – reach and loud environments
Comparing humans and VUIs, the speech recognition systems are lacking millennia of contextual experience and VUIs still encounter challenges when trying to understand the semantics of a sentence

The solution

Having analyzed the data provided, the following project design was suggested and, later on, implemented.

  • Establish the exact text samples best suiting the requirements
  • Divide the existing shoppers base by segments according to gender, age and geographical position to provide the most diverse and accurate samples with population’s characteristics considered
  • Extend the shoppers base for the needed segments coverage
  • Record 2, 600 voice sample with people reading pieces of text out loud with further uploads directly to the system
  • Implement bug reporting for all mistakes encountered in the client’s software to additionally calibrate the app released for crowd sourcing

Program’s aim:

Supply the system with variety of quality assured samples and exact wording

Ensure the accuracy of texts voicing

Calibrate the system for particular languages

Provide the needed data for successful product launch ​

The outcome

The client was provided with the ongoing supplying of new and diverse data which resulted in successful calibration of AI with the required languages. By the end of the project, the product was tested and released on the respective markets.

Important to highlight – due to remaining challenges of the VUIs, taking your product to the new level would require deep segmentation and constant supply of additional data for numerous linguistical features consideration.

Stop Losing Your Clients Via Digital Channels
Get Free Demo