Course introduction

Forensic phonetics, forensic linguistics, and this course

Dr Massimiliano Canzi

Introduction

Did you know: the superb lyrebird of south-eastern Australia is one of the best vocal mimics known. About 80% of a male’s song is made of imitations of other species, mostly other birds, and the imitations are remarkably accurate.

Sources: Wildlife Victoria; Dalziell et al. (2012), Animal Behaviour

Who I am

I am Dr Massimiliano Canzi, Max for short.

I studied linguistics in Milan (BA), Manchester (MA) and York (MSc), and returned to Manchester for my PhD.

I work on how listeners perceive speech and how speakers produce it, mostly with experiments and quantitative methods. Forensic speech science is one of my interests.

Who are you?

Tell us briefly:

Your name and your programme
How much phonetics you have done so far
A forensic case involving speech or language that you have heard of, in the news, in a podcast or on TV
One thing you would like to take away from this course

The course

Did you know: female superb lyrebirds also sing and mimic, especially when defending their nests, when they imitate predators and alarm calls. Their songs were long overlooked: adult females look almost identical to young males, so singing females were easily taken for males.

Source: Dalziell & Welbergen (2016), Frontiers in Ecology and Evolution

How the course works

Seminar, [day and time]: one chapter of the book per session, with discussion and exercises
Practical sessions with Praat, later in the semester: measuring speech yourself
Office hours: [day and time], by appointment
Ilias: announcements, readings and submissions

Assessment

Component	Weight
[component]	[weight]
[component]	[weight]

Details follow on Ilias.

The book

Leemann, A., Perkins, R., Sullivan Buker, G., & Foulkes, P. (2024). An Introduction to Forensic Phonetics and Forensic Linguistics. Routledge.

Written by practitioners and researchers in both fields
Real cases, exercises and suggestions for further reading in every chapter
We cover the forensic phonetics part (chapters 1 to 7), two sessions on forensic linguistics, and chapter 13 on expert witnesses

Syllabus

Chapters of Leemann et al. (2024):

	Session	Chapter
1	Course introduction	1
2	Phonetic analysis	2
3	Speaker profiling	3
4	Speaker comparison	4
5	Earwitness evidence	5
6	Authentication, enhancement and speech content	6
7	Linguistic analysis in the asylum procedure	7
8	Language in the legal process	10
9	Authorship analysis	11, 12
10	Expert witnesses	13
11	Praat practical 1: measuring your own voice	2, 4
12	Praat practical 2: casework with the class recordings	3–5

Slides and materials

The slides are written in R Markdown and shown with reveal.js. Each week you get the slides (.html) and the readings.

The figures use simulated data: realistic numbers, made up for teaching. The code behind every figure is in the .Rmd file, at github.com/mcanzi/ws27_fss, if you are curious. You do not need R for this course.

Exercises

The book has exercises in every chapter, and we do many of them together.

Forensic questions often have no single correct answer, so we discuss how you reached yours and how sure you can be
Try each exercise before you read the solution

Recording your voice

For the Praat practicals (sessions 11 and 12), each of you records yourself once in the LingLab booth, at any time before session 11. It takes about 15 minutes.

Item	Content	File name
Word list	19 words in the frame Say ___ again, read three times	`S07_words.wav`
Sentences A	three sentences, read normally	`S07_A.wav`
Sentences B	three other sentences, read normally	`S07_B.wav`
Disguise	sentences A again, in a disguised voice	`S07_A_disguise.wav`

Use the ID you receive in place of S07, and upload the four files to the course folder.

The word list

Read each word in the frame Say ___ again, for example Say heed again:

heed	hid	head	had
hard	hod	hoard	hood
who’d	heard	hud	think
this	three	car	nurse
water	butter	little

Read the whole list three times, at your normal speed. The first eleven words differ only in the vowel; the others contain sounds that vary across accents of English.

Sentences A and B

Set A	Set B
The north wind blew hard across the dark water all night.	In the dark, the water looked cold and much deeper than it was.
My sister thinks the red car parked by the school is hers.	Nobody thinks a car like that belongs in such a quiet street.
We heard the news on the radio and called two friends at once.	They heard a loud noise and ran back to the house.

Read each set at your normal speed, as one short text.

The disguised reading

Read set A once more, in a voice that you think would stop your classmates from recognising you. Choose your own strategy, for example a different pitch, a different accent, a creaky or nasal voice. Do not whisper: we need f0
Write down your strategy on the form you receive, and do not tell your classmates

What this course is, and is not

This course gives you a grounding in the methods and problems of forensic speech science, and can be a first step towards further study. Forensic phoneticians train for years, at postgraduate level and in forensic labs.

It does not qualify you to do casework, and the book’s authors say the same about their book.

Forensic phonetics and linguistics

Did you know: a lyrebird raised in captivity in New South Wales in the 1920s is said to have learnt flute tunes and passed them on to wild lyrebirds after its release. Flute-like songs can still be heard there, but recordings made over 40 years suggest that the local song has stayed stable over that period.

Sources: Powys, Taylor & Probets (2013), Environmental Humanities; Probets, Taylor & Powys (2020), Corella

Two fields

Forensic phonetics deals with speech and voices as evidence: who is speaking and what was said, or whether a recording has been edited
Forensic linguistics deals with language as evidence, from the authorship of a text (session 9) to the language of police interviews and courtrooms (session 8)

Despite the names, few practitioners work in both. Most specialise further, so casework is often done by teams.

Where the fields meet

Language analysis in the asylum procedure (LAAP): where does an asylum seeker’s speech suggest they grew up? (session 7)
Record keeping, report writing and acting as an expert witness: the same principles apply in both fields, and in all forensic sciences (session 10)

You do it every day

When you answer the phone, you analyse the first few words, usually without noticing:

Do I know this voice?
Does it match the name on the screen?
Roughly how old is the caller, and do they sound local?
Do they sound calm?

Forensic phonetics asks similar questions, with systematic methods and an estimate of how reliable each answer is.

An old idea: shibboleth

And the Gileadites took the passages of Jordan before the Ephraimites: and it was so, that when those Ephraimites which were escaped said, Let me go over; that the men of Gilead said unto him, Art thou an Ephraimite? If he said, Nay; Then said they unto him, Say now Shibboleth: and he said Sibboleth: for he could not frame to pronounce it right.

Judges 12:5–6, King James Bible

A crude speaker profile: /ʃ/ for Gileadites, /s/ for Ephraimites.

A recent shibboleth

In 2022, Ukrainians reportedly asked strangers to say паляниця (palianytsia, a kind of bread) to unmask Russian saboteurs (Hyde, 2022)
The word was believed to be hard to pronounce for speakers of Russian
Shibboleth itself is now an everyday word for any sign of group membership

Discussion

Is the pronunciation of one word enough to tell where someone is from?

What else could explain the “wrong” pronunciation?
What if the speaker knows the test?
What happens if the test gets it wrong?

No: a single word is not a sound basis for a legal decision, and the book makes the same point. Session 3 covers how profiling is actually done.

No two samples are the same

Simulated vowels: the same speaker never says a vowel twice the same way, and different speakers overlap.

What the book covers

Chapter	Question
3 Speaker profiling	What can a voice tell us about the speaker?
4 Speaker comparison	Do two recordings come from the same speaker?
5 Earwitness evidence	Can a witness recognise a voice they heard?
6 Authentication, enhancement, content	Is the recording genuine? What was said?
7 LAAP	Where did an asylum seeker grow up?
13 Expert witnesses	How do experts report, and what do courts do with it?

Beware the CSI effect

Did you know: lyrebirds are famous for imitating chainsaws and other machines. But the well-known recordings come from captive birds, such as one at Adelaide Zoo: as of 2014, no recording was known of a wild lyrebird mimicking a mechanical sound.

Source: Taylor (2014), The Conversation

Forensic science on TV

Experts collect samples, interview suspects and analyse everything, alone
Results arrive in seconds, from impressive software
Conclusions are a match or a mismatch, “with 99.4% certainty”
A “voiceprint” on screen matches a stored sample perfectly

Forensic science in reality

Analyses take days
Experts rarely, if ever, report “match” or “mismatch”: there is almost always some doubt
Objectivity and transparency matter more than speed
No two samples of speech are ever identical, so there is no such thing as a voice “match”

Even DNA is not a “match”

A DNA sample from a crime scene is identical to the suspect’s DNA
This does not mean the suspect committed the crime: it means they could have
So could everyone else who shares that profile, possibly thousands of people
Other evidence, such as where the suspect was or a motive, is for the court to weigh

Two propositions

The expert weighs the evidence under two competing propositions:

Prosecution: the two samples come from the same source
Defence: the two samples come from different sources

The expert then asks how much more likely the evidence is under one proposition than under the other. More in sessions 4 and 8.

Similar is not enough

Simulated mean F0 of adult male speakers. Solid: offender. Dashed: suspect.

Similar is not enough

In both cases, the offender and the suspect differ by only 2 Hz. But:

Case 1: about 20% of speakers have a mean F0 within 5 Hz of the offender’s
Case 2: only about 2.1% do

The 2 Hz difference is much stronger evidence in case 2, because so few speakers have such a low F0. How common a feature is in the population is called its typicality.

Discussion

A TV expert says: “The voices match, with 99.4% certainty.”

What could the 99.4% possibly mean, and what would you need to know to judge it?
Who should decide whether the suspect is guilty?

Working with forensic data

Did you know: while courting, a male superb lyrebird imitates a whole mobbing flock: the alarm calls of several species harassing a predator. He does it when a female tries to leave and during mating, although no predator is around: an acoustic illusion that may keep her from leaving.

Source: Dalziell et al. (2021), Current Biology

The material

Forensic material ranges from the mundane (phone calls about bank transfers) to the distressing (threats, recordings of crimes)
Practitioners describe three main strains: graphic content, pressure, and cases that touch on personal experience (Giménez, Elstein & Queralt, 2020)
In this course we only work with the cases discussed in the book

Offenders and victims

Notorious offenders are widely remembered; their victims much less so
In 2019, New Zealand’s Prime Minister Jacinda Ardern refused to name the man who attacked two mosques in Christchurch
Forensic analyses often focus on the person who produced the speech, often the offender: the book asks readers to keep the people harmed in view, and to choose their words with care

Ethics

Know the limits of your expertise and of your methods, and never comment beyond them
Be objective: no conclusion stronger or weaker than the analysis supports
The expert’s duty is to the court, whichever side asked for the report
The accused is presumed innocent: if the evidence leaves doubt, the court must know

Further resources

IAFPA: the International Association for Forensic Phonetics and Acoustics
IAFLL: the International Association for Forensic and Legal Linguistics
International Journal of Speech, Language and the Law
Nolan, McDougall & Hudson (eds.) (in press), The Oxford Handbook of Forensic Phonetics
Forensic Transcription Australia: Helen Fraser’s site on transcripts of covert recordings

Summary

Forensic phonetics treats speech as evidence; forensic linguistics treats language as evidence
No two speech samples are identical, so experts weigh similarities and differences and do not declare a “match”
Similarity counts only together with typicality
The expert assists the court, objectively, within the limits of their expertise
Cases involve real people, victims included, so we describe them with care

References

Leemann, A., Perkins, R., Sullivan Buker, G., & Foulkes, P. (2024). An Introduction to Forensic Phonetics and Forensic Linguistics. Routledge. Chapter 1.
Giménez, R., Elstein, S., & Queralt, S. (2020). The pandemic and the forensic linguistics caseworker’s wellbeing: Effects and recommendations. International Journal of Speech, Language and the Law, 27(2), 233–254. https://doi.org/10.1558/ijsll.19548
Hyde, L. (2022, February 26). Saboteurs spark suspicion and solidarity in Kyiv. Politico.

Sources of the animal facts

Dalziell, A. H., Maisey, A. C., Magrath, R. D., & Welbergen, J. A. (2021). Male lyrebirds create a complex acoustic illusion of a mobbing flock during courtship and copulation. Current Biology, 31(9), 1970–1976. https://doi.org/10.1016/j.cub.2021.02.003
Dalziell, A. H., & Welbergen, J. A. (2016). Elaborate mimetic vocal displays by female superb lyrebirds. Frontiers in Ecology and Evolution, 4, 34. https://doi.org/10.3389/fevo.2016.00034
Dalziell, A. H., Welbergen, J. A., Igic, B., & Magrath, R. D. (2012). Fooling the experts: Accurate vocal mimicry in the song of the superb lyrebird. Animal Behaviour, 83, 1401–1410. https://doi.org/10.1016/j.anbehav.2012.03.009
Powys, V., Taylor, H., & Probets, C. (2013). A little flute music: Mimicry, memory, and narrativity. Environmental Humanities, 3, 43–70. https://doi.org/10.1215/22011919-3611230
Probets, C., Taylor, H., & Powys, V. (2020). A sonographic analysis of a flute-like dialect in territorial songs of the superb lyrebird Menura novaehollandiae in the New South Wales North Coast and New England Tableland bioregions. Corella.
Taylor, H. (2014). Lyrebirds mimicking chainsaws: Fact or lie? The Conversation.
Wildlife Victoria. Superb lyrebird. https://wildlifevictoria.org.au/learn/native-species-post/superb-lyrebird/

Slides and code

These slides

mcanzi.de/ws27_fss/01_course_introduction.html

Code and all sessions

github.com/mcanzi/ws27_fss