WEBVTT

1
00:00:00.200 --> 00:00:02.350
- The best part about
being a software engineer

2
00:00:02.350 --> 00:00:03.890
is that things are always changing.

3
00:00:03.890 --> 00:00:05.890
Nothing ever stands still.

4
00:00:05.890 --> 00:00:07.020
I'm William Walker.

5
00:00:07.020 --> 00:00:08.550
I'm a software engineer here at the BBC

6
00:00:08.550 --> 00:00:09.700
on the Voice and AI team,

7
00:00:09.700 --> 00:00:11.470
and I am one of the developers responsible

8
00:00:11.470 --> 00:00:12.770
for creating Beeb.

9
00:00:12.770 --> 00:00:15.117
I've always been interested
in seeing how things work.

10
00:00:15.117 --> 00:00:16.170
When I was younger,

11
00:00:16.170 --> 00:00:17.700
I spent a lot of time out on the bike

12
00:00:17.700 --> 00:00:20.300
and just seeing model
engines and things like that.

13
00:00:20.300 --> 00:00:21.680
Cycling and software,

14
00:00:21.680 --> 00:00:25.500
they both require a lot of
upkeep, a lot of maintenance.

15
00:00:25.500 --> 00:00:26.380
It can be difficult,

16
00:00:26.380 --> 00:00:28.990
but it can be worth it to
really put the time in.

17
00:00:28.990 --> 00:00:31.880
Beeb is a voice assistant from the BBC,

18
00:00:31.880 --> 00:00:34.020
and it allows people to
interact with us verbally

19
00:00:34.020 --> 00:00:35.750
rather than through a
more traditional method,

20
00:00:35.750 --> 00:00:38.470
such as a mouse or a touchscreen.

21
00:00:38.470 --> 00:00:40.810
The voice assistant
field is very different

22
00:00:40.810 --> 00:00:43.370
from typical projects
that we might've worked on

23
00:00:43.370 --> 00:00:44.420
in the past.

24
00:00:44.420 --> 00:00:46.250
Azure has been really good for us.

25
00:00:46.250 --> 00:00:48.750
In the early stages,
it was really fun just

26
00:00:48.750 --> 00:00:51.830
to have free reign, if you
like, to just experiment

27
00:00:51.830 --> 00:00:54.210
with the different Azure
technologies and just learn

28
00:00:54.210 --> 00:00:55.190
and just see how things work.

29
00:00:55.190 --> 00:00:58.260
It provides all the tools
that we need in one place

30
00:00:58.260 --> 00:00:59.980
to create a branded voice assistant

31
00:00:59.980 --> 00:01:03.290
and making the unique
experiences that we want to.

32
00:01:03.290 --> 00:01:04.210
Being a software engineer,

33
00:01:04.210 --> 00:01:06.350
you're just one person
in a much larger team,

34
00:01:06.350 --> 00:01:08.010
and so you work with
a wide range of people

35
00:01:08.010 --> 00:01:09.740
with a wide range of skill sets

36
00:01:09.740 --> 00:01:11.910
to solve a wide range of problems.

37
00:01:11.910 --> 00:01:13.750
We've built Beeb because we
know there's growing demand

38
00:01:13.750 --> 00:01:15.900
for people to access
content and their services

39
00:01:15.900 --> 00:01:17.130
using their voice.

40
00:01:17.130 --> 00:01:19.960
We want to have greater
control over our brands

41
00:01:19.960 --> 00:01:22.580
and create a more robust
connection with our audience.

42
00:01:22.580 --> 00:01:25.470
So our ambitions with Beeb can't be met

43
00:01:25.470 --> 00:01:27.720
by just developing for
existing voice assistants

44
00:01:27.720 --> 00:01:28.553
on the market.

45
00:01:28.553 --> 00:01:30.580
With this one, we saw some challenges

46
00:01:30.580 --> 00:01:32.150
around how we create the wake words

47
00:01:32.150 --> 00:01:34.392
and how we collect the data
to enable that to happen.

48
00:01:34.392 --> 00:01:37.810
The U.K. has a wide range of accents,

49
00:01:37.810 --> 00:01:39.580
and so creating Beeb, it's important to us

50
00:01:39.580 --> 00:01:41.540
that it can be used by everyone.

51
00:01:41.540 --> 00:01:43.530
And that means that we
have spent a lot of time

52
00:01:43.530 --> 00:01:45.430
developing the wake words to make sure

53
00:01:45.430 --> 00:01:47.010
that it can be understood by

54
00:01:47.010 --> 00:01:48.970
as wide range of accents as possible.

55
00:01:48.970 --> 00:01:50.930
Our assistant uses Azure Bot Services,

56
00:01:50.930 --> 00:01:53.240
which is where it's all
set up and configured.

57
00:01:53.240 --> 00:01:55.210
We use a number of Cognitive Services,

58
00:01:55.210 --> 00:01:57.900
such as LUIS for language understanding.

59
00:01:57.900 --> 00:02:01.350
We use the Custom Neural
Voice and the custom keywords,

60
00:02:01.350 --> 00:02:02.300
which is "Okay, Beeb",

61
00:02:02.300 --> 00:02:04.160
and we have QnA Maker

62
00:02:04.160 --> 00:02:07.230
for simpler question
and answer interactions.

63
00:02:07.230 --> 00:02:11.170
The Beeb client app is a UWP
app that runs on Windows.

64
00:02:11.170 --> 00:02:14.080
It uses the Azure Bot Services
Direct Line Speech channel,

65
00:02:14.080 --> 00:02:15.800
which is a web socket connection,

66
00:02:15.800 --> 00:02:18.420
specifically designed
for voice interaction,

67
00:02:18.420 --> 00:02:20.000
to connect to our assistant,

68
00:02:20.000 --> 00:02:22.920
and that's configured with
our custom text to speech,

69
00:02:22.920 --> 00:02:25.360
our custom ASR model, the custom voice.

70
00:02:25.360 --> 00:02:27.200
Within our assistant, we rely on a number

71
00:02:27.200 --> 00:02:30.290
of internal BBC services
to retrieve content.

72
00:02:30.290 --> 00:02:32.350
We also have our own analytics platform.

73
00:02:32.350 --> 00:02:36.113
So if a user was to say,
"Okay, Beeb, play Radio One".

74
00:02:37.038 --> 00:02:39.310
- [Beeb] Here's BBC Radio One.

75
00:02:39.310 --> 00:02:41.260
- We're only just getting
started with Beeb,

76
00:02:41.260 --> 00:02:43.210
and so we're just laying the foundations

77
00:02:43.210 --> 00:02:44.720
for our next phase of work,

78
00:02:44.720 --> 00:02:46.930
and we're really ambitious
in terms of platform

79
00:02:46.930 --> 00:02:47.940
and feature support.

80
00:02:47.940 --> 00:02:52.857
It's a really exciting time to
be a part of the department.