02/14/2019 Learned Model submission page opened.
02/21/2019 Stage2 page opened.
This competition takes 2 stages to decide winners. Please check "Procedure of the competition" at the bottom of this page and submit the learned model of your last posted result by the end of stage 1 from following "learned model submission page".
Recently, applications utilizing voice user interface are becoming more and more popular. In those applications, there is a need to make much more context aware responses by considering user profiles. To this end, NTT has organized this competition, called NTT corevo (*1) challenge. In this competition, we ask you to identify speakers’ ages and genders from short voice data (a few seconds).
(*1) NTT Group has adopted the unified brand name of “corevo” for all of its initiatives that utilize the various AI-related technologies that the group has developed through its R&D
In this competition, NTT asks you to create a model which classifies voice data recorded by smart phones into 6 classes (gender x age).
Male, female : 2 classes
Children (8 to 12 ages, before voice breaking), adult (20 to 50 ages), elderly adult (over 65 ages) : 3 classes
Training data consist of 37,806 samples. For the stage 1 evaluation, there are 17,888 samples. The average length of voice data is about 6 seconds and sampling frequency is 16 KHz.
Procedure of the competition
This competition takes 2 stages to decide winners.
- stage 1
As normal competitions, we ask you to create a model and submit a prediction result for test data (data A) from the start of the competition to the end date of stage 1. Use score board as the guide, create generalized model.
Note that you must submit a model itself as well which you selected, by the end of the stage 1. You can submit a model using the submission page from February onwards. You don’t need to submit a model each time, but you can submit multiple models as well. In this case, your last submission will be regarded as your final model. Be sure that you cannot submit a model after the end of the stage 1.
- stage 2
To evaluate an ability of generalization of your model, we will publish new data (data B). You need to submit prediction result for data B by using a model you eventually selected. We will decide final standings by this prediction result.Caution : You shouldn’t change/modify a model which you selected in the stage 1. If you do it, you will be disqualified with the competition. So, please be careful especially when you deal with random numbers.