Effects of Reverberation on Conversations in Rooms

1
Effects of Reverberation on Conversations in Rooms
2
Well-established speechsegregation cues

• Differences in voice pitch
• Differences in spatial location

1
3
Well-established speechsegregation cues

• Differences in voice pitch
• Differences in spatial location

1
4
The beneficial effect of difference in the
fundamental frequency

• What has been done up to now ? Brokx Nooteboom
  (1984) Summerfield Culling (1994) Culling,
  Hodder and Toh (2003)
• First experiment
• Perspectives

5
The beneficial effect of difference in the
fundamental frequency

• What has been done up to now ? Brokx Nooteboom
  (1984) Summerfield Culling (1994) Culling,
  Hodder and Toh (2003)
• First experiment
• Perspectives

6
Brokx Nooteboom (1984) Monotonized sentences.
target
F0 (Hz)
masker
F0 difference (Hz)
0 3 6 9 20
100
The higher the difference in F0, the smaller the
percentage of errors.
F0 difference (semitones)
7
Brokx Nooteboom (1984) Monotonized sentences.
target
F0 (Hz)
masker
F0 difference (Hz)
0 3 6 9 20
100
The higher the difference in F0, the smaller the
percentage of errors, except when this difference
in F0 is one octave (inseparability of the
harmonics of target and interfering speech)
F0 difference (semitones)
8
Brokx Nooteboom (1984) Monotonized sentences.
target
F0 (Hz)
masker
F0 difference (Hz)
0 3 6 9 20
100
Even with the same F0 intelligibility is not
completely lost (60 errors)
F0 difference (semitones)
9
Brokx Nooteboom (1984) real speech, either
intonated or deliberately monotone
target
F0 (Hz)
masker
Monotone speech small effect of difference in F0
(not perfectly monotonous pronunciation?)
Natural speech large, consistent effect of
difference in F0
10
Summerfield and Culling (1994) Synthesized,
vibrato vowels.
F0 difference (Hz)
0 12 26
0
-5
Beneficial effect of difference in F0 on the
identification of vowels
-10
Identification threshold (dB)
-15
-20
0 2 4
F0 difference (semitones)
11
Culling, Summerfield and Marshall (1994) Effect
of reverberation on concurrent vowel
identification.
(A)
(B)
(C)
(D)
12
Concurrent vowel identification in
reverberation as a function of vibrato depth.
in reverberation
Vibrato Depth
F0 Difference
(semitones)
(semitones)
0.5
2
0.125
0.25
0.5
1
2
13
Culling, Hodder and Toh (2003) Running speech.
14
Culling, Hodder and Toh (2003) Running speech.
Does reverberation destroy the listeners ability
to use prosodic information conveyed by the
intonation contour ?
15
Culling, Hodder and Toh (2003) Running speech.
Reverberation affects the intonated voice more
than the monotonous one, but the intonated speech
is intrinsically more intelligible.
Reverberation has little effect on prosodic
information in the intonation contour.
16
The results Culling Hodder and Toh were difficult
to interpret due to the effect of role F0
contours in intelligibility. In order to avoid
influences on intelligibility, weve proposed to
use completely artificial F0 contours on running
speech.
Original
Monotonised
Vibrato
17
The beneficial effect of difference in the
fundamental frequency

• What has been done up to now ? Brokx Nooteboom
  (1984) Summerfield Culling (1994) Culling,
  Hodder and Toh (2003)
• First experiment
• Perspectives

18
First experiment
F0 Difference
Vibrato Depth
(semitones)
(semitones)
0
1
0
0.5
1
2
No spatial separation
Modulation frequency 1Hz
19
First experiment
F0 Difference
Vibrato Depth
(semitones)
(semitones)
0
1
0
0.5
1
No spatial separation
2
Modulation frequency 5 Hz
20
Modulation Frequency 1Hz
Speech Reception Thresholds
21
Modulation Frequency 5Hz
Speech Reception Thresholds
22
Modulation Frequency 5Hz
Interferer the pearl was worn in a thin silver
ring
Speech Reception Thresholds
23
Modulation Frequency 1Hz
Interferer the pearl was worn in a thin silver
ring
Speech Reception Thresholds
24
Conclusion

• The results obtained by Culling, Summerfield and
  Marshall (1994) regarding the double vowels
  identification seem to predict some of the
  aspects of the present results regarding real
  speech.
• When the modulation frequency is 1 Hz we can
  still exploit F0 differences in reverberation.

We can still employ a rF0 in reverberation as
long as the modulation width lt 2 rF0
A modulation frequency of 5Hz is already too hard
for the listener to track.
25
The beneficial effect of difference in the
fundamental frequency

• What has been done up to now ? Brokx Nooteboom
  (1984) Summerfield Culling (1994) Culling,
  Hodder and Toh (2003)
• First experiment
• Perspectives

26
Now..?

• Does reverberation blur the harmonicity of the
  target, or the masker, or both ?

Next experiment will tell us
27
Next experiment
Vibrato Depth (blue interfering sentence)
(semitones)
2
0
Modulation frequency 5 Hz
Difference F0 2 semitones
28
Next experiment
Vibrato Depth (blue interfering sentence)
(semitones)

• Reverberation distorts intelligibility by
  blurring the harmonicity of
• the interferer ?

2
0
Modulation frequency 5 Hz
Difference F0 2 semitones
29
Next experiment
Vibrato Depth (blue interfering sentence)
(semitones)

• Reverberation distorts intelligibility by
  blurring the harmonicity of
• the target ?
• the interferer ?

2
0
Modulation frequency 5 Hz
Difference F0 2 semitones
30
Next experiment
Vibrato Depth (blue interfering sentence)
(semitones)

• Reverberation distorts intelligibility by
  blurring the harmonicity of
• both ?
• the target ?
• the interferer ?

2
0
Modulation frequency 5 Hz
Difference F0 2 semitones
31
Culling, Summerfield and Marshall (1994) Effect
of reverberation on concurrent vowel
identification.
Vibrato Depth 2 semitones
Modulation frequency 5 Hz

• The cue which enables low SRT is the harmonic
  relationship between maskers and targets or only
  masker, but not simply the one of the target !

Extrapolate these results to the natural
speechleft.
32

• Thanks for your attention.