Using Creaky Voice Index in Forensic Phonetics – Is it valid and is it reliable?

Using Creaky Voice Index in Forensic Phonetics –

Is it valid and is it reliable?

____________________________Tuija Niemi-Laitinen

Forensic Scientist/Technical DepartmentCrime Laboratory

National Bureau of Investigation

Finnish PoliceNational Bureau of Investigation

Creaky voice…

• Is a voice of an “expert”• Sounds more mature• Sounds more “relaxed” (not much muscle action)• Reflects “cool” attitude - ” -• Is a social marker of “educated speech” in Finland?• Is a marker of Swedish speaking person in Finland?• When female speakers try to imitate male voice?• Signals uncertainty and/or nervousness

• F-072


Creaky voice is thought to involve high adductive tension and medial compression, but little longitudinal tension



Creak phonation

• is produced with vibrating vocal folds but at a very low frequency

• pitch has been observed to be extremely low, and would appear to be controlled by aerodynamic factors and not by varying the longitudinal tension (like other qualities, Ní Chasaide & Gobl 1999).


Creak phonation

• the F0 and amplitude of consecutive glottal pulses is further known to be very irregular

• because of the high adductive tension, only the ligamental part of the vocal folds is vibrating

• the folds are relatively thick and compressed, and the ventricular folds may also be somewhat adducted, so that their inferior surfaces come in contact with the superior surfaces of the true vocal folds

• this would thus create an even thicker vibrating structure


Creak phonation• the mean airflow rate has been observed to be

very low (Ní Chasaide & Gobl 1999)• the resulting low tension and heavy vibrating

mass are responsible for the slower and irregular vibration

• both subglottal pressure and the glottal airflow are lowered compared to modal phonation

• creak is produced at a flow rate of 12-20 cc/s while pulses are produced in a frequency range from 25 to 50 Hz (male) and 50-130 (female).


Potentially functional, idiolectal or emotional use of creaky voice

• Lexically distinctive feature in some languages• Laryngealisation before vowel onset• Word boundary marker between two vowels (/-V#V-/)• Filler in filled pauses• Utterance or turn final preboundary marker• Occurring only in unstressed vowels in some

speakers• Overall idiolectal feature in some speakers(Iivonen, Nordic Prosody 2004)


50.00 100.00 150.00 200.00 250.00 300.00

F0

0

100

200

300

400

500

600F

req

ue

nc

y

Mean = 129.2032Std. Dev. = 39.18816N = 5,084

F0 distribution, speaker 072, female, spontaneous speech (F0 range set to 70-300 Hz)


Study of creaky voice index

• How much creakiness there is among Finnish female speakers in both reading and spontaneous speech

• 33 female speakers

• Age 18-61 (mean 36.1)

• Speech samples were recorded via GSM phones and stored on a computer


Age of the speakers

0

10

20

30

40

50

60

70

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33

Speaker

Yea

rs


Analysis

• For the F0 analysis I used the Praat program

• Manually-set analysis range (70-300 Hz)

• Praat program was used to create PitchTiers

• SPSS program was used to calculate the statistics and F0 histograms from the PitchTiers


Creaky voice limits for female speakers

• 140 Hz (Moosmüller 2001)

• 100 Hz ("Finn-voice-limit")


RESULTS

• Results show the percentages how often a certain speaker used creaky voice

• Results show the difference in using creaky voice between the spontaneous speech and text reading tasks intra-individually

• Inter-individual differences can also be seen



F0 Statistics with SPSS, speaker F072, spontaneous speech, with F0 script

NValid 3844

Missing0

Mean 141.7840

Median 151.6740

Mode 62.62(a)

Std. Deviation 33.83130

Variance 1144.557

Skewness -1.009

Std. Error of Skewness .039

Kurtosis -.076

Std. Error of Kurtosis .079

Range 146.53

Minimum 62.62

Maximum 209.16

Percentiles 10 77.7722


NValid 5084

Missing0

Mean 129.2032

Median 145.9571

Mode 69.98(a)

Std. Deviation 39.18816

Variance 1535.712

Skewness -.277

Std. Error of Skewness .034

Kurtosis -1.188

Std. Error of Kurtosis .069

Range 225.00

Minimum 69.98

Maximum 294.98

Percentiles 10 75.1251

F0 Statistics with SPSS, speaker F072, spontaneous speech, F0 analysis range 70-300


50,00 100,00 150,00 200,00 250,00 300,00

F0

0

100

200

300

400

500

600

700

Fre

qu

en

cy

Mean = 175,237Std. Dev. = 48,827N = 4 987

F0 histogram, speaker F93, spontaneous speech


50,00 100,00 150,00 200,00 250,00 300,00

F0

0

100

200

300

400

500

600

700

Fre

qu

en

cy

Mean = 196,395Std. Dev. = 42,03707N = 7 819

F0 histogram, speaker F93, text reading


Hypothesis-1

• Some female speakers use creaky voice (less than 140 Hz) almost 50% of the total amount of the measured F0 points.

• Answer-1: This study shows that the minimum percentage was 2.4, maximum 81.3 and average 18.7 with 33 speakers in spontaneous speech.

• With the text reading task, the values were 0.7, 80 and 13.8%, respectively.


Hypothesis-2

• About one half of the Finnish female speakers use creaky voice (less than 140 Hz) at least 10% of the total amount of measured F0 points

• Answer-2: The result in this study show that 21 out of total 33 female speakers (63.6 %) used creaky voice at least 10% of the total amount of measured F0 points


Hypothesis-3

• All Finnish female speakers have measured points less than 100 Hz.

• Answer-3: Yes.

• Minimum percentage was with the speaker no. 88=0.8, the maximum value was with the speaker no. 72=34.9, and the average was 9.2%.

• For nine speakers out of 33, the percentage was over 10%.


Hypothesis-4• Spontaneous speech is creakier than read speech.

• Answer-4: Yes.

• Only two speakers showed the opposite tendency

• Nine out of 33 speakers had a very small difference (less than 1%) between these two speech styles (limit 140 Hz)

• With the limit of 100 Hz the results were as follows:

• Read speech was creakier only with four speakers

• Nine out of 33 speakers had a very small difference (less than 1%) between these two speech styles


Hypothesis-5

• If the average F0 value is over 200 Hz, the CVI is 1-2% (limit 140 Hz).

• Answer-5: In this study there are five speakers whose F0 average is over 200 Hz with spontaneous speech, and five with read speech. This hypothesis is true only for two of them with spontaneous speech and three with read speech.


Hypothesis-6

• If the average F0 value is less than 160 Hz, the CVI is over 20% (limit 140 Hz).

• Answer-6: In this study there are eight speakers whose F0 average is less than 160 Hz (spontaneous speech)

• This hypothesis is true for seven out of eight persons with spontaneous speech and none with read speech.


CONCLUSION

• The speakers differ greatly from each other in respect with the use of creaky voice. The percentages vary from 0.2 to over 80%

• There is also a huge variation between the speakers in the way they use creaky voice (only in the final falls of utterances, final falls of every sentence, in every word final, throughout the speech)


VALIDITY• Validity of the measurements is a relevant question in respect

with the extent in which the use of creaky voice signals uncertainty or nervousness of the speaker

• This question is relevant also in forensic phonetics• When a suspect is being recorded, does his/her voice get

creaky because s/he knows much about the case (cf. the idea of lie detectors) or that s/he is uncertain or just nervous?

• If a person's voice is creaky due to uncertainty and/or excitement, it is definitely an area that has to be studied in forensics due to the fact that suspects often are under severe stress when they are accused of committing a crime (if they are guilty, of course)

• Then the reference speech samples should deviate from the incriminating speech samples in the level of stress and uncertainty


RELIABILITY• All the recordings were made via GSM phones• The samples were band-pass filtered without the

actual fundamental frequency values• The auto-correlation method for F0

measurements still can be used for measuring using the upper harmonics

• This can be checked by inspecting the spectrogram

• Reliability is going to be checked by analysing simultaneous GSM and microphone recordings


COMPARISON QUALITY• Questioned sample(s)• ( ) large amount of personal features • ( ) small amount of personal features• ( ) few words (words_______)• ( ) too short sample• ( ) poor technical quality • Reference samples• ( ) natural variation of the speech is well represented • ( ) natural variation of the speech is poorly represented • ( ) not verbatim text_________• ( ) poor technical quality _________


COMPARISON QUALITY• Comparison quality• ( ) good• ( ) inadequate• Similarities• ( ) no significant similarities• ( ) are concerned to be idiolectal speech features• ( ) are concerned to be ordinary speech features• Other: • ( ) different speech style • ( ) time between the recordings• ( ) different channels of the recordings • Possible voice disguise• ( ) not recognized• ( ) recognized Q ( ) R ( )


DIFFERENCES

• ( ) no significant differences

• ( ) significant differences

• ( ) do not fit within the range of natural variation observed in the reference sample

• Estimation of differences

• ( ) time between the recordings

• ( ) physical/psychological state

• ( ) voice disguise

• ( ) medicine/alcohol/drugs

• ( ) speech situation

Using Creaky Voice Index in Forensic Phonetics – Is it valid and is it reliable?

Documents

male voice

hz finnvoice

study of creaky voice

spontaneous speech

creaky voiceresults

pitchtierscreaky voice

resulting low tension

vibrating vocal folds