CONVERGENCE at The 26th Interspeech Conference in Rotterdam, The Netherlands

Kalle

CONVERGENCE’s Kalle Lahtinen presented results on “Investigating Affect Mining Techniques for Annotation Sample Selection in the Creation of Finnish Affective Speech Corpus” by Kalle Lahtinen, Einari Vaaras, Liisa Mustanoja and Okko Räsänen.

CONVERGENCE Doctoral researcher Kalle Lahtinen took part in the 26th Interspeech conference, which was held August 17 – 21, 2025, in Rotterdam. The theme of the conference for 2025 was “Fair and Inclusive Speech Science and Technology”. Interspeech is an annual scientific conference organized by the International Speech Communication Association (ISCA). Interspeech conferences include papers covering all the scientific and technological aspects of speech.

Kalle

During the conference, Kalle presented a poster describing the work of Lahtinen, Vaaras, Mustanoja and Räsänen titled “Investigating Affect Mining Techniques for Annotation Sample Selection in the Creation of Finnish Affective Speech Corpus”. The accepted conference paper was also the first peer-reviewed article to be included in Kalle’s doctoral dissertation. In their work, they described the first affective speech corpus based on spontaneous Finnish speech (FinnAffect), which will be made public through the Language Bank of Finland. The main findings of the presented work were related to the methods for automatically finding affectively diverse speech samples (i.e speech samples with differing emotional or affective content) from large-scale, unannotated speech data to be used in human evaluation. One of the major challenges in studying affective or emotional language today is related to the nature of available datasets: most emotional speech corpora contain acted speech, recorded in controlled environments spoken by few individuals. FinnAffect aims to provide new resources for studying affective language in the context of everyday, spontaneous expression, with speech from thousands of different speakers.  The work was received with interest by many attendees of the conference, underlining the need for acquiring realistic and natural speech data for speech research.

In addition to presenting their work, Kalle also attended workshops on “Extracting insights from your complex data: interpretable statistical methods in speech science” and “Interpretability Techniques for Speech Models”as well as a plethora of oral presentations, poster events and social gatherings, offering concrete ideas on how to move forward with the study on affective Finnish as well as new collaboration opportunities.

The conference offered a landscape view on the status of speech science and technology, introducing work from basic linguistic research to state-of-the-art speech applications and everything in between.