1
00:00:41,230 --> 00:00:43,800
On completion of this training sequence,

2
00:00:43,800 --> 00:00:49,510
you will be able to use your voice
as a tool to produce high quality subtitles.

3
00:00:49,510 --> 00:00:53,540
In particular,
you are going to learn about how to use it

4
00:00:53,540 --> 00:00:57,430
to support flawless dictation
through voice projection,

5
00:00:57,430 --> 00:01:01,690
pacing, articulation and voice modulation.

6
00:01:02,650 --> 00:01:05,350
This is the agenda of this presentation.

7
00:01:11,360 --> 00:01:15,680
Voice projection
is strictly related to breathing.

8
00:01:15,680 --> 00:01:17,230
It is the use of voice

9
00:01:17,230 --> 00:01:22,050
in a way that the person
to whom you are talking listens to you.

10
00:01:22,050 --> 00:01:25,650
When respeaking,
you need to talk to a machine.

11
00:01:25,650 --> 00:01:29,840
In normal talking,
we use the upper part of our lungs,

12
00:01:29,840 --> 00:01:35,750
while in respeaking we need to use
belly breathing as automatically as possible,

13
00:01:35,750 --> 00:01:39,640
so as to be able and respeak on the long run.

14
00:01:39,640 --> 00:01:40,660
This means

15
00:01:40,660 --> 00:01:45,280
that your voice projection should be regular
and well-balanced,

16
00:01:45,280 --> 00:01:49,340
as it is aimed at excellent use
of voice commands

17
00:01:49,340 --> 00:01:53,930
in a manner that the machine
can consistently listen to your voice

18
00:01:53,930 --> 00:01:55,150
as you trained it.

19
00:01:56,370 --> 00:01:57,530
To do this,

20
00:01:57,530 --> 00:02:02,670
you need to keep practicing belly breathing
and speaking at the same time.

21
00:02:02,670 --> 00:02:04,850
When speaking or respeaking,

22
00:02:04,850 --> 00:02:09,340
try to see if you manage to reduce
the number of recognition errors

23
00:02:09,340 --> 00:02:12,900
by avoiding ups and downs in the volume.

24
00:02:12,900 --> 00:02:17,190
In particular,
pay attention to thinking all sentences

25
00:02:17,190 --> 00:02:22,140
as if they needed a raising tone;
articulating properly;

26
00:02:22,140 --> 00:02:28,880
avoiding sudden accelerating your dictation,
and avoiding ups and downs in your dictation,

27
00:02:28,880 --> 00:02:32,740
but more importantly
avoid talking to the machine

28
00:02:32,740 --> 00:02:37,030
the way you talk to humans,
as in this demo of mine.

29
00:02:37,030 --> 00:02:41,840
I am not a native speaker of English,
so don't focus on the way I pronounce words.

30
00:02:41,840 --> 00:02:46,460
Rather, try to imitate
the way I project my voice.

31
00:03:04,350 --> 00:03:08,340
Pacing is the rate at which
each of us speaks.

32
00:03:08,340 --> 00:03:12,670
In public speaking, they teach you to vary
your speaking speed.

33
00:03:12,670 --> 00:03:14,150
Avoid doing this.

34
00:03:14,150 --> 00:03:16,400
As we have seen in the previous section,

35
00:03:16,400 --> 00:03:20,490
you need to talk to the machine
the most robotic way possible.

36
00:03:20,490 --> 00:03:23,460
This allows for a good speech recognition.

37
00:03:23,460 --> 00:03:26,360
Pacing plays an important role in respeaking,

38
00:03:26,360 --> 00:03:30,490
as it is your capacity to speak clearly
at a given speech rate.

39
00:03:30,490 --> 00:03:35,700
To understand how quickly you can talk,
try and shadow the TV news,

40
00:03:35,700 --> 00:03:39,170
a parliamentary speech
or a fast-rate webinar.

41
00:03:39,170 --> 00:03:45,110
In respeaking, pacing has to be adapted
to the speaker’s rate constantly.

42
00:03:50,980 --> 00:03:54,280
Articulation is strictly related to pacing

43
00:03:54,280 --> 00:03:59,390
as you should not only speak quickly
to keep the same pace as the speaker.

44
00:03:59,390 --> 00:04:02,760
You should also speak
in a clear and distinctive manner,

45
00:04:02,760 --> 00:04:06,950
so that the machine can transcribe
the words we have in mind

46
00:04:06,950 --> 00:04:09,820
and not words that sound similar.

47
00:04:09,820 --> 00:04:14,740
To do this, keep an eye
on the articulation rules in your language.

48
00:04:14,740 --> 00:04:20,350
This is extremely important to reduce
recognition errors while respeaking.

49
00:04:20,350 --> 00:04:25,560
To do so, it is sometimes necessary
to change the way you pronounce a word

50
00:04:25,560 --> 00:04:30,510
through macros
or other peculiar ways of articulating words,

51
00:04:30,510 --> 00:04:33,520
as per Element number 4.

52
00:04:33,520 --> 00:04:39,820
Always keep track of special articulation
so as to make the most out of it.

53
00:04:40,550 --> 00:04:43,480
As a rule of thumb,
valuable for many languages,

54
00:04:43,480 --> 00:04:47,110
remember to always articulate
all syllables in a word

55
00:04:47,110 --> 00:04:50,610
and all words in a sentence
the way they should,

56
00:04:50,610 --> 00:04:57,080
and not as you do it in your daily life,
if this changes the way they are pronounced.

57
00:04:57,080 --> 00:05:01,930
With polycentric languages like English,
this also means you need to check

58
00:05:01,930 --> 00:05:08,100
if the automatic speech recognition software
that you use can be adjusted to your accent.

59
00:05:08,100 --> 00:05:11,570
In this case,
be consistent with this standard

60
00:05:11,570 --> 00:05:17,210
and do not change the way you articulate
depending on the accent of the speaker.

61
00:05:17,210 --> 00:05:21,500
Also, avoid stressing syllables too much,

62
00:05:21,500 --> 00:05:25,920
as a machine is not a human being.

63
00:05:25,920 --> 00:05:29,150
In case of possibly ambiguous words,

64
00:05:29,150 --> 00:05:33,310
pronounce the word
without stressing a given syllable too much

65
00:05:33,310 --> 00:05:38,960
as in many languages vowel length
can make the difference between two words,

66
00:05:38,960 --> 00:05:41,990
as in “shot” and “short”.

67
00:05:49,680 --> 00:05:53,770
Finally, voice modulation is the use
you make of your voice

68
00:05:53,770 --> 00:05:56,580
in terms of volume and tone.

69
00:05:56,580 --> 00:06:02,290
In human conversation you adapt it
to the message or effect you want to convey.

70
00:06:02,290 --> 00:06:06,410
In respeaking, this means
to be consistent with the three aspects

71
00:06:06,410 --> 00:06:09,840
we have dealt with in the previous sections

72
00:06:09,840 --> 00:06:13,640
and keep modulation as robotic as possible.

73
00:06:13,640 --> 00:06:18,260
This may seem easy, but it can be a challenge
with emotional speakers,

74
00:06:18,260 --> 00:06:20,800
as you may tend to imitate them.

75
00:06:20,800 --> 00:06:24,790
If you realise
the number of recognition errors increases,

76
00:06:24,790 --> 00:06:28,160
a wrong voice modulation may be the reason.

77
00:06:34,560 --> 00:06:37,630
In this video lecture,
we have seen how important it is

78
00:06:37,630 --> 00:06:43,440
to make a good use of your voice
while breathing, so as to support dictation.

79
00:06:43,440 --> 00:06:46,800
In particular, we have seen
how to command voice projection,

80
00:06:46,800 --> 00:06:53,070
pacing, articulation and voice modulation
to avoid as many recognition errors

81
00:06:53,070 --> 00:06:56,570
while trying to subtitle
challenging speakers.