1
00:00:00,00 --> 00:00:06,02
(gentle music)

2
00:00:06,02 --> 00:00:07,08
- [Instructor] Now that you have learned

3
00:00:07,08 --> 00:00:11,04
how to do machine learning audio classification,

4
00:00:11,04 --> 00:00:14,09
I have an exciting challenge for you.

5
00:00:14,09 --> 00:00:17,04
I thought you have to use Teachable machine

6
00:00:17,04 --> 00:00:19,05
with google.com to do a demo.

7
00:00:19,05 --> 00:00:22,01
So you can refer back to the same demo lesson

8
00:00:22,01 --> 00:00:24,02
or you can also look at the handout

9
00:00:24,02 --> 00:00:27,01
where I have given the steps for this.

10
00:00:27,01 --> 00:00:28,08
So here's what you're going to do.

11
00:00:28,08 --> 00:00:30,01
We talked about

12
00:00:30,01 --> 00:00:33,06
so many different potential HJI use cases

13
00:00:33,06 --> 00:00:39,01
in real work environment and one of them was the ability

14
00:00:39,01 --> 00:00:44,05
for an AI model to understand water dripping sound.

15
00:00:44,05 --> 00:00:49,03
It could be a pipe leaking our water overflowing,

16
00:00:49,03 --> 00:00:51,08
and that sound could be a separate class

17
00:00:51,08 --> 00:00:55,01
that you can train right here on Teachable machine.

18
00:00:55,01 --> 00:00:57,01
So that's what we are going do.

19
00:00:57,01 --> 00:00:58,09
So I would like you

20
00:00:58,09 --> 00:01:01,01
to record background noise

21
00:01:01,01 --> 00:01:05,04
just like we did in the demo as one class.

22
00:01:05,04 --> 00:01:10,03
And you can use the mic here and just stay quiet

23
00:01:10,03 --> 00:01:12,06
and record the background noise

24
00:01:12,06 --> 00:01:15,05
and make it generate samples for you.

25
00:01:15,05 --> 00:01:19,06
The second class, you can make a similar one to the bell

26
00:01:19,06 --> 00:01:21,02
so that we have three classes.

27
00:01:21,02 --> 00:01:22,07
One is going to be the background noise,

28
00:01:22,07 --> 00:01:25,02
one is going to be another class like bell

29
00:01:25,02 --> 00:01:26,08
or any sound that you want.

30
00:01:26,08 --> 00:01:29,09
And the third is going to be the dripping water sound.

31
00:01:29,09 --> 00:01:34,00
And what did we do with the bell? I literally clicked mic.

32
00:01:34,00 --> 00:01:37,05
The mic was ready to listen and said add sample

33
00:01:37,05 --> 00:01:41,07
and I took a bell and made some noise.

34
00:01:41,07 --> 00:01:43,04
So you could do something like that.

35
00:01:43,04 --> 00:01:46,03
You can use a bell or if you don't have a bell handy,

36
00:01:46,03 --> 00:01:48,05
you can find any other sound

37
00:01:48,05 --> 00:01:51,08
but consistent sound so that you can get some samples.

38
00:01:51,08 --> 00:01:55,03
And we want the sample sizes to be close, okay?

39
00:01:55,03 --> 00:01:58,02
So it cannot be like a hundred samples of bell

40
00:01:58,02 --> 00:02:01,00
and you know, 20 samples of water or something like that

41
00:02:01,00 --> 00:02:04,02
because the more the number of samples,

42
00:02:04,02 --> 00:02:05,08
the model is going to learn.

43
00:02:05,08 --> 00:02:08,08
So we don't want to bias the model.

44
00:02:08,08 --> 00:02:13,01
So the second sound is bell or some other song.

45
00:02:13,01 --> 00:02:17,07
Similarly, in this challenge, you add a third class

46
00:02:17,07 --> 00:02:20,02
and you can call it water, you can call it any name.

47
00:02:20,02 --> 00:02:22,01
The class name doesn't actually have to say

48
00:02:22,01 --> 00:02:25,04
what the sample is about, but I like to just call it water.

49
00:02:25,04 --> 00:02:28,05
So I'll call it water. Add a third class.

50
00:02:28,05 --> 00:02:33,02
And now actually take water in a cup, take some spoon,

51
00:02:33,02 --> 00:02:36,04
and create that pipe leak sound that you want

52
00:02:36,04 --> 00:02:39,00
to emulate from your environment

53
00:02:39,00 --> 00:02:41,07
and collect some samples.

54
00:02:41,07 --> 00:02:44,07
Similarly, it'll collect the sample

55
00:02:44,07 --> 00:02:48,06
and then it'll extract the one second samples from there.

56
00:02:48,06 --> 00:02:52,05
And your sample set, three classes will all be ready.

57
00:02:52,05 --> 00:02:55,09
And then what do you do? Do you remember from the demo?

58
00:02:55,09 --> 00:03:00,00
You simply press model, train model.

59
00:03:00,00 --> 00:03:04,01
You keep the AP science acid is and done

60
00:03:04,01 --> 00:03:06,04
and you will be able to preview the model.

61
00:03:06,04 --> 00:03:09,01
Now you can test the model with the sound

62
00:03:09,01 --> 00:03:12,01
of water dripping and then you're done.

63
00:03:12,01 --> 00:03:13,07
And if you want, you can export it

64
00:03:13,07 --> 00:03:15,02
like we saw in another exercise.

65
00:03:15,02 --> 00:03:18,09
If you want to take this to another mobile application

66
00:03:18,09 --> 00:03:22,01
or integrate this programmatically anywhere else,

67
00:03:22,01 --> 00:03:24,08
then you have the option of generating the code,

68
00:03:24,08 --> 00:03:27,07
doing everything from the expert model option.

69
00:03:27,07 --> 00:03:28,06
That's it.

70
00:03:28,06 --> 00:03:32,04
So good luck and think about the use case.

71
00:03:32,04 --> 00:03:35,07
Think about the use case of where the water is dripping,

72
00:03:35,07 --> 00:03:38,05
where the pipe is leaking, what would you do?

73
00:03:38,05 --> 00:03:41,04
And I will give you the solution in the next video.

74
00:03:41,04 --> 00:03:43,00
Good luck.