3 Papers Accepted at INTERSPEECH 2025

May 27, 2025 | News

We’re glad to share that the three papers, co-authored by our Machine Translation Group at FBK, have been accepted at INTERSPEECH 2025!

Echoes of Phonetics: Unveiling Relevant Acoustic Cues for ASR via Feature Attribution. Dennis Fucci, Marco Gaido, Matteo Negri, Mauro Cettolo, Luisa Bentivogli.
How to Connect Speech Foundation Models and Large Language Models? What Matters and What Does Not. Francesco Verdini*, Pierfrancesco Melucci*, Stefano Perna*, Francesco Cariaggi*, Marco Gaido, Sara Papi, Szymon Mazurek, Marek Kasztelnik, Luisa Bentivogli, Sébastien Bratières, Paolo Merialdo, Simone Scardapane.
Granary: Speech Recognition and Translation Dataset in 25 European Languages. Nithin Rao Koluguri, Monica Sekoyan, George Zelenfroynd, Sasha Meister, Shuoyang Ding, Sofia Kostandian, He Huang, Nikolay Karpov, Jagadeesh Balam, Vitaly Lavrukhin, Yifan Peng, Sara Papi, Marco Gaido, Alessio Brutti, Boris Ginsburg.

MT Group at FBK Follow

#MachineTranslation Research Unit @FBK_research. #nlproc #deeplearning #ai

Avatar MT Group at FBK @fbk_mt ·

19 Jun

Our pick of the week by @DennisFucci: "Speech Representation Analysis Based on Inter- and Intra-Model Similarities" by Yassine El Kheir, Ahmed Ali, and Shammur Absar Chowdhury (ICASSP Workshops 2024)

#speech #speechtech

Dennis Fucci @DennisFucci

Findings from https://ieeexplore.ieee.org/document/10669908 show that speech SSL models converge on similar embedding spaces, but via different routes. While overall representations align, individual neurons learn distinct localized concepts.
Interesting read! @fbk_mt

Reply on Twitter 1935711333431037957 Retweet on Twitter 1935711333431037957 2 Like on Twitter 1935711333431037957 3 Twitter 1935711333431037957

Retweet on Twitter MT Group at FBK Retweeted

Avatar Fondazione Bruno Kessler - FBK @fbk_research ·

10 Jun

Cosa chiedono davvero gli italiani all’intelligenza artificiale?
FBK in collaborazione con RiTA lancia un’indagine aperta a tutte/i per capire usi reali, abitudini e bisogni.

Bastano 10 minuti per partecipare, scopri di più: https://magazine.fbk.eu/it/news/italiani-e-ia-cosa-chiediamo-veramente-allintelligenza-artificiale/

Reply on Twitter 1932368425910734864 Retweet on Twitter 1932368425910734864 6 Like on Twitter 1932368425910734864 12 Twitter 1932368425910734864

Retweet on Twitter MT Group at FBK Retweeted

Avatar Marco Gaido @mgaido91 ·

11 Jun

🚀 Last call for the Model Compression for Machine Translation task at #WMT2025 (co-located with #EMNLP2025)!
Test data out on June 19 ➡️ 2 weeks for evaluation!
Can you shrink an LLM and keep translation quality high?
👉 https://www2.statmt.org/wmt25/model-compression.html #NLP #ML #LLM #ModelCompression

Reply on Twitter 1932810353974346104 Retweet on Twitter 1932810353974346104 9 Like on Twitter 1932810353974346104 8 Twitter 1932810353974346104

Avatar MT Group at FBK @fbk_mt ·

11 Jun

Our pick of the week by @beomseok_lee_: "ALAS: Measuring Latent Speech-Text Alignment For Spoken Language Understanding In Multimodal LLMs" by Pooneh Mousavi, @yingzhi_wang, @mirco_ravanelli, and @CemSubakan (2025)

#SLU #speech #multimodal #LLM

Beomseok LEE @beomseok_lee_

Speech-language models show promise in multimodal tasks—but how well are speech & text actually aligned? 🤔

This paper https://arxiv.org/abs/2505.19937 proposes a new metric to measure layer-wise correlation between the two, with a focus on SLU tasks. 🔍🗣️📄

Reply on Twitter 1932717374731661474 Retweet on Twitter 1932717374731661474 2 Like on Twitter 1932717374731661474 5 Twitter 1932717374731661474