aggregates performance metrics for each keyword to generate single report for overall summary

[ljj7975]

performance metric captured include (units for time is ms)

• mfcc generation time (avg, max, min)
• inference time (avg, max, min)
• accuracy

As part of development I have ran it with 110 data samples (100 positive, 10 negative) result I have for that is

• accuracy : 99 / 100 (fails to capture negative case - it's known issue)
• inferenceTimeAvg : 38.55272727034224
• inferenceTimeMax : 126.9000000320375
• inferenceTimeMin : 33.10000000055879
• mfccGenTimeAvg : 45.685454544780605
• mfccGenTimeMax : 103.59999997308478
• mfccGenTimeMin : 33.39999995660037

[ljj7975]

btw, that was on chrome with my mac. I guess the mfcc generation / inference time will change based on the device & version / type of browser

[ljj7975]

this is result from firefox on my mac

• accuracy: 0.9 (99/110)
• inferenceTimeAvg: 21.60909090909091
• inferenceTimeMax: 560
• inferenceTimeMin: 9
• mfccGenTimeAvg: 41.95454545454545
• mfccGenTimeMax: 106
• mfccGenTimeMin: 27

Glad to see some difference in inference time

[ljj7975]

I am currently writing code for display result on the page I guess it should take long and once we have that, we can try it on the tablet as well