Header

Suche
Institut für Computerlinguistik
Volker Dellwo

Volker Dellwo, Prof. Dr.

  • Associate Professor
  • Phonetics & Speech Sciences

Electronic IDs: Google Scholar,OrcID,LinkedIn; Download CV: CV

My research investigates how speaker identity is encoded, perceived, and modeled in human and machine speech processing. I work at the intersection of voice biometrics, forensic phonetics, and machine learning, with a particular focus on the challenges posed by synthetic speech and deepfake technologies. I lead forensic voice research at the University of Zurich as Chair of the Centre for Forensic Phonetics and Acoustics(CFPA) and serve as co-director of  JPFrench International(JPFi) , where we deliver advanced forensic voice biometric services in high-stakes legal contexts. I am a co-founder of the Voice Communcation Sciences Association (VoCSA) and an active contributor to the EU MSCA Doctoral Network project Voice Communication Sciences  (VoCS), helping to shape the European research and training landscape in this emerging field. Together with international collaborators, I founded and co-organised the inaugural VoiceID conference in Zurich in 2022, that is now an interdisciplinary forum, bringing together researchers and practitioners working on voice identity across neuro-psychology, phonetics, computer sciences and AI with applications in medical and forensic sciences. As chair of the steering committee of the Linguistic Research Infrastructure (LiRI), I contribute to the development and provision of research technologies for our community.

My research is strongly supported by an excellent and highly engaged team of PhD candidates:  Alessandro De Luca,  Alessandra Metzger,Andrea Fröhlich,Aref Farhadi Pour,David Manuel Grünert, Erdem Baha Topbas, Masoumeh Chapariniya, Ming Jin, Sadia Azad, Timothy Tianze Xu

Look at ourPhonetics & Speech Sciences courses and check out my speech & voice tools on the LiRI Resource Hub Gitbook 

Key Projects

Weiterführende Informationen

VOCS - A Doctoral Network in Communication Sciences

Mehr zu VOCS - A Doctoral Network in Communication Sciences

Time period: 2024 - 2028

Funding: Marie Skłodowska-Curie Actions

Collaborators: Elisa Pellegrino, Srikanth Madikeri, Pascale Belin, Tomi Kinnunen, Junichi Yamagishi, Yue Zhang

Read more: VOCS - A Doctoral Network in Communication Sciences

NCCR - Evolving Language

Mehr zu NCCR - Evolving Language

Time period: 2020 - 2028

Funding: Swiss National Science Foundation (SNSF)

Collaborators: Alexis Hervais-Adelman

Read more: NCCR - Evolving Language

FluSS - Fluency in Speech and Signing

Time period: 2026 - 2030

Funding: Swiss National Science Foundation (SNSF)

Collaborators: Sarah Ebling, Tobias Haug, Nivia de Jong

Read more: FluSS - Fluency in Speech and Signing

Recent Publications

Weiterführende Informationen

T-SNE scatterplot illustrating how the 11 call types from the updated vocal repertoire cluster together based on similarity in their acoustic structure. Each point in the scatterplot represents a call and the different colours of the points depict the different call types.

An updated vocal repertoire of wild adult bonobos (Pan paniscus)

Mehr zu An updated vocal repertoire of wild adult bonobos (Pan paniscus)

Research on bonobo and chimpanzee vocal behaviour has grown, yet most work relies on unvalidated qualitative call repertoires. Quantitative data are essential for understanding species-wide communication and for comparisons across primates. We provide the first quantitative validation of a Pan vocal repertoire using wild bonobo calls. Analysing more than 1500 calls from 53 adults over 33 months with random-forest models, we identify 11 acoustically distinct call types. We also discuss challenges in defining vocal repertoires from field data and suggest future methods for capturing the graded complexity of bonobo vocal communication.

Example utterance after post-processing steps. Tiers from top to bottom: words, phones, sentence, truth, trusted, score. Spectrogram settings: view range = 0Hz-6000Hz, window length = 0.05s, dynamic range = 90.0 dB.

NumberLie: a game-based experiment to understand the acoustics of deception and truthfulness

Mehr zu NumberLie: a game-based experiment to understand the acoustics of deception and truthfulness

We introduce the NumberLie game, a method for collecting natural deceptive speech with precise ground-truth labels and immediate consequences for lying. The setup enables isolated, simultaneous audio recording of multiple players in a controlled lab or online environment as they engage in a number-based deception game. We detail the technical features that ensure accurate truth–lie classification and real-time outcomes, supported by performance-based financial incentives. The design is highly adaptable, allowing researchers to manipulate conditions while minimising extraneous variability.

A Density plots showing distributions of mean f0 values for younger and older, as well as male and female speakers. Dashed lines represent group means. B Boxplots showing mean, range and interquartile range of mean f0 values per speaker. ‘f_old’—older female speakers, ‘f_yng’— younger female speakers, ‘m_old’—older male speakers, ‘m_yng’—younger male speakers. 1, 2, 3, 4, 5—individual speaker IDs

Listeners are biased towards voices of young speakers and female speakers when discriminating voices

Mehr zu Listeners are biased towards voices of young speakers and female speakers when discriminating voices

Own-age advantages are well documented in face recognition, but it is unclear whether they occur in voice recognition. Younger (n=42) and older adults (n=32) completed a same/different speaker discrimination task involving younger and older speakers of both sexes. Sensitivity (d′) showed no speaker–listener age or sex interactions. Instead, younger listeners outperformed older listeners, and male voices were discriminated more accurately than female voices. Response bias (c) indicated a stronger “same” bias for younger and female speakers. These results inform models of voice identity processing and have implications for forensic contexts involving earwitness judgments.

All publications

You can find all publications here. 

Educational Videos

  • was-ist-phonetik-thumbnail

    Was ist Phonetik

    Video abspielen
  • presenter_pro_compressed_thumbnail

    PresenterPro: How to prompt participants with Praat

    Video abspielen
  • databaseExplorer_compressed_thumbnail

    DatabaseExplorer: How to edit a speech corpus with Praat

    Video abspielen
  • ema_tr_compressed_thumbnail

    Electromagnetic Articulography: How to measure the movements of articulators

    Video abspielen

Weiterführende Informationen

Office Address

AND-2.38

Andreasstr. 15
CH-8050 Zurich
Switzerland

Map

Phone: +41(0)44 634 2995

Email

Phone or office appointments

Please book phone or office appointments with the admin office:

Tamar Tolcachier
Email

Master Thesis in Phonetics

If you are interested in a MA thesis in Phonetics & Speech Sciences, please fill in this form.

Find my office in the building

Take Lift B to the second floor.

At the glass door on the right, you’ll find a phone with my number next to it (4 27 75).

Please ring me and I will come to pick you up at the door.

If someone lets you through the glass door, turn right and walk down the corridor. My office is the third from last on the left.

Email contact

Due to the high volume of incoming emails, my response rate is currently limited. If you do not receive a reply within a week, it is unlikely that I will respond. I apologise for any inconvenience this may cause. For urgent matters, please contact:
- Alexandra Buenzli (buenzli@cl.uzh.ch) – Department of Computational Linguistics
- Claudia Leibundgut (claudia.leibundgut@uzh.ch) – administrative, personnel, and financial matters
- Elisa Pellegrino (elisa.pellegrino@uzh.ch) – Phonetics & Speech Sciences Group and CFPA