ling 411 - simon fraser universitymjmunro/courses/l411h07.pdf · ling 411: assignment 2 comment...

16
10/13/16 1 LING 411 Topics in Applied Phonetics www.sfu.ca/~mjmunro/courses/l411.html “Voiceprints” ? • term is not used by forensic phoneticians • sometimes used by others to refer to spectral representations of speech (mainly spectrograms) wrongly conveys the impression of something comparable to a fingerprint

Upload: others

Post on 29-May-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

1

LING411TopicsinAppliedPhonetics

www.sfu.ca/~mjmunro/courses/l411.html

“Voiceprints”?

• term is not used by forensic phoneticians

• sometimes used by others to refer to spectral representations of speech (mainly spectrograms)

• wrongly conveys the impression of something comparable to a fingerprint

Page 2: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

2

Acoustic representations of speech are notcomparable to fingerprints

Speech has organic and learned components.

Fingerprints have no learned component.

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

organic component of speech:(mostly outside the speakers control)

size of SLVTproperties of vocal folds

nature of soft tissuessome vocal abnormalities

Page 3: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

3

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

organic component of speech:(mostly outside the speakers control)

identical twins è very similar vocal tracts, except for in-utero factors (phenotypic)

fraternal twins; other siblings è as with fingerprints, similarities are likely to exist

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

learned component of speech:

(acquired through socialization early in life;mostly unconscious; strong social influences)

Page 4: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

4

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

learned component of speech:

dialect: mainly affects vowels & rhoticity in English

UK – many varietiesUS

AustraliaNZ

Canada

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

learned component of speech:

dialectparticular choices of social variants (tɑkɪŋ; tɑkɪn; tɑkin)

use of articulators (bunched tongue /ɹ/ ?)voice quality choices

pitch variability; speech rate

identical twins: even when raised together may learn differently (Nolan & Oh, 1996)

Page 5: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

5

Acoustic representations of speech are notcomparable to fingerprints

organic - learned dichotomy

learned component of speech:

identical twins: even when raised together may learn differently

e.g., in UK study,

one twin used word-initial /ɹ/; the other used /ʋ/(labio-dental approximant)

(Nolan & Oh, 1996)

Overlapping influences

Organic factors influencepitch capabilities

vocal tract resonances (formants)voice quality

ANDLearned aspects of speech affect

use of pitchuse of resonance

voice quality..

Page 6: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

6

Overlapping influences

Organic and learned factors influence many of the SAME aspects of speech

BUT

In terms of measurement, we have no means of fully separating the organic effects on speech

from the learned effects.

Overlapping influences

not an issue for fingerprints because they are organic

Page 7: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

7

Intra-speaker variability

unconscious and not purposeful: constant variability in nature of segmental and prosodic phenomena

purposeful: we vary the way we speak for social reasons (solidarity, respect, authority)

purposeful: U may employ vocal disguise

Nolan & Grigoras

obscene phone call recordings

ABCD

+ K

these are U voices, but not established for certain that A = B = C = D

Page 8: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

8

appropriateness of /ɪ/ for measurements? (WCE)

How do these figures capture dynamic vowel properties?

/aʊ/

/oʊ/

Page 9: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

9

1937 Hz tone, total duration > 25 seconds, 4 different episodesformant tracker followed this tone; not F2

Three options:

erase 25 seconds (decrease the voice duration)

manually correct F2 values

filter the noisy fragments (using a notch filter)

Page 10: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

10

Speech with loud 1 kHz tone

The appropriate notch filter is a bandstop filter that removes the energy at 1000 Hz.

1000 Hz component

Speech with loud air conditioner noise

Page 11: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

11

RE: F0, K is sotto voce (low intensity), which might result in lower F0

NB: Automatic formant tracker uses 25 Hz bins

Page 12: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

12

LTS Comparison

• Questions for Assignment 2 are now posted

• New reading assigned: Schilling & Marsters

LING 411: Assignment 2

Comment briefly as indicated, and include your responses along with your box plots.

1. In a couple of sentences describe the Mean F0 data (female and male), commenting on the medians and nature of the distributions.

2. For the female jitter data only, identify one thing that the three measures have in common and one way in which they differ.

3. For the female shimmer data only, identify one thing that the three measures have in common and one way in which they differ.

4. What do you think the multiple SFO measures (3 for “Rainbow” and 4 for “Boy”) indicate about the reliability of the SFO data?

Page 13: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

13

More terminology & concepts from Nolan & Grigoras

SNR

What is it?What causes a poor SNR?

isolexemic events

labile (ˈleɪˌbaɪl) nature of F0

unintentional increase in F0 for speech produced in noise (Lombard reflex)

Page 14: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

14

Background

• threatening phone message (in English) on answering machine

• speaker had a Cantonese accent• suspect, K = Mr. Lo (charged)• Rogers was contacted by defense lawyer

Q: Does K = U?

Speech material

Rogers collected varied sample from K

• 3 readings of the threatening message (isolexemic!)• standard reading passage• wordlist• casual conversation

Why?

Page 15: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

15

Speech material

Rogers collected varied sample from K

• 3 readings of the threatening message (isolexemic!)• standard reading passage• wordlist• casual conversation

Why? Mainly to explore consistency in Lo’s speech

Cantonese accent in English

1. Cantonese speakers often don’t produce syllable-final [ɹ] as in car

Findings: both speakers often omitted final [ɹ]The voice on the tape produced the word “terminator” twice

Tape PRONOUNCED the r in both cases Lo never pronounced ɹ in final position

2. Cantonese speakers often use glottal stop [ʔ] to replace final voiceless stops [p t k] or sometimes glottalize voiceless stop (i.e., ʔp, ʔt, ʔk)

Lo used [ʔ] alone much more often than the tape did. Lo also said [neʔ] for “name” whereas the tape said [nem].

3. Cantonese speakers sometimes have trouble with [s] and [ʃ], espeically if they occur near each other.

Tape had much better control over these sounds than did Lo.

Page 16: LING 411 - Simon Fraser Universitymjmunro/courses/L411H07.pdf · LING 411: Assignment 2 Comment briefly as indicated, and include your responses along with your box plots. 1. In a

10/13/16

16

/ fɪntʃ steɪʃn̩ /

for Lo: ntʃst è / nst /

CONCL: Tape showed fewer of the characteristics of a Cantonese accent in English than did Lo.

If U were really Lo, Lo would have had to “reduce his accent” while making the threatening call. [Highly improbable!]