@fbk_mt profile picture
MT Group at FBK @fbk_mt
Joined October 2015
  • Tweets 3,281
  • Following 456
  • Followers 1,398
419 Photos and videos
We're excited to share that three papers from our lab have been accepted at #Interspeech2026 !! 🍾 #SpeechTranslation #SpeechAI #NLProc #FBK #Interspeech
2
6
8
155
2️⃣ "Better Late Than Never: Evaluating Latency Metrics for Simultaneous Speech-to-Text Translation" πŸ‘₯ Peter Polak, @sarapapi, @luisabentivogli, OndΕ™ej Bojar πŸ“„ arxiv.org/abs/2509.17349

Better Late Than Never: Meta-Evaluation of Latency Metrics for...

Simultaneous speech-to-text translation systems must balance translation quality with latency. Although quality evaluation is well established, latency measurement remains a challenge. Existing...

arxiv.org
2
2
62
MT Group at FBK@fbk_mt
Jun 11
3️⃣ "Cross-Attention is Half Explanation in Speech-to-Text Models" πŸ‘₯ @sarapapi, @DennisFucci, @mgaido91, @negri_teo, @luisabentivogli πŸ‡ͺπŸ‡Ί DVPS EU project πŸ“„ arxiv.org/abs/2509.18010

Cross-Attention is Half Explanation in Speech-to-Text Models

Cross-attention is a core mechanism in encoder-decoder architectures, widespread in many fields, including speech-to-text (S2T) processing. Its scores have been repurposed for various downstream...

arxiv.org
2
2
47
MT Group at FBK@fbk_mt
Jun 11
1️⃣ "Do What I Say: A Spoken Prompt Dataset for Instruction-Following" πŸ‘₯ @MaikeZufle, @sarapapi, Fabian Retkowski, Szymon Mazurek, @mkasztelnik, Alexander Waibel, @luisabentivogli, @_janius_ πŸ‡ͺπŸ‡Ί Meetween EU project πŸ“„ arxiv.org/abs/2603.09881

Do What I Say: A Spoken Prompt Dataset for Instruction-Following

Speech Large Language Models (SLLMs) have rapidly expanded, supporting a wide range of tasks. These models are typically evaluated using text prompts, which may not reflect real-world scenarios...

arxiv.org
3
2
80
MT Group at FBK@fbk_mt
Jun 10
Our pick of the week by @lina_conti : "Greater accessibility can amplify discrimination in generative AI" by @CarolinHolterm, @minhducbui_nlp, @KaitlynZhou, @vjhofmann, @kelina1124, @anne_lauscher πŸ“° arxiv.org/abs/2603.22260 #GenderBias #SpeechLLM

Greater accessibility can amplify discrimination in generative AI

Hundreds of millions of people rely on large language models (LLMs) for education, work, and even healthcare. Yet these models are known to reproduce and amplify social biases present in their...

arxiv.org
Lina Conti@lina_conti
Jun 10
Pick of the week @fbk_mt: "Greater accessibility can amplify discrimination in generative AI" Gender bias in speech-based LLMs examined from multiple angles: a user survey, automatic bias measurement, and pitch manipulation experiments. arxiv.org/pdf/2603.22260
3
6
888
Late update, but we had two great talks last month! #MachineTranslation #FBK #NLProc #GenderBias #SpeechSynthesis
1
1
156
From Harm Lameris, a PhD candidate at @KTHuniversity on "Communicative Functions of Synthesized Speech: Modelling Prosody and Voice Quality Dynamics"
1
125
Janiça Hackenbuchner, a PhD candidate at Ghent University on "Contextual Cues and Gender Ambiguity in MT: From Human Perception to Model Interpretability"
1
2
79
MT Group at FBK@fbk_mt
Jun 3
Our pick of the week by @dhairya_su47605 : "Scaling Laws for Precision" by @tanishqkumar07, Zachary Ankner, @bfspectorShiekh, @blake__bordelon, @Muennighoff, @mansiege, @CPehlevan, Christopher RΒ΄e, @AdtRaghunathan πŸ“°arxiv.org/abs/2604.19565 #Quantization #LLM #ScalingLaw

Detecting Hallucinations in SpeechLLMs at Inference Time Using...

Hallucinations in Speech Large Language Models (SpeechLLMs) pose significant risks, yet existing detection methods typically rely on gold-standard outputs that are costly or impractical to obtain....

arxiv.org
Dhairya Suman@dhairya_su47605
Jun 3
Pick of the week @fbk_mt Super interesting paper on the limitations of quantization, demonstrating how post-training quantization scales poorly in data. arxiv.org/abs/2411.04330
1
4
170
⭐ For our #PickOfTheWeek, this paper explores an important question for modern speech AI: πŸŽ™οΈ Which Evaluation for Which Speech Model? πŸ‘₯ Authors: @Maureendss , @EeshanDhekane Speech foundation models are evolving rapidly, but evaluation practices are still fragmented.
2
2
167
more replies
πŸ‘‰πŸ» What is being evaluated πŸ‘‰πŸ» Which capabilities are required πŸ‘‰πŸ» What task/protocol constraints define the setup One key takeaway is that benchmark choice can strongly shape conclusions about model performance and generalization.
1
83
✨ A useful read for anyone working on speech models, multimodal AI, and benchmarking. #PickOfTheWeek #MTUnit #FBK #Speech
1
1
44
🏝️ Yesterday at #LREC2026, Palma de Mallorca! @lina_conti presented "Voice, Bias, and Coreference: An Interpretability Study of Gender in Speech Translation" at the poster session. πŸ“„Paper: arxiv.org/abs/2511.21517 πŸ’»Code: github.com/lina-conti/voice-… #SpeechTranslation #NLProc

Voice, Bias, and Coreference: An Interpretability Study of Gender...

Unlike text, speech conveys information about the speaker, such as gender, through acoustic cues like pitch. This gives rise to modality-specific bias concerns. For example, in speech translation...

arxiv.org
1
6
299
🏝️ Also at #LREC2026, Palma de Mallorca! @luisabentivogli presented our second paper. "Phonetic-based Ranking for Improved Pseudo-Labeling in Low-Resource ASR" πŸ“„ Paper: eloquenceai.eu/wp-content/up… Bravi!! πŸŽ‰
1
1
132
How does the granularity of speech-text pairs impact SpeechLLM performance, and what is the optimal way to interleave tokens? Furthermore, what are the best practices for generating synthetic data to boost training?🧐
1
3
197
The paper "Data-Centric Lessons To Improve Speech-Language Pretraining" by @vishaal_urao, Zhiyun Lu, Xuankai Chang, Yongqiang Wang, Violet Z. Yao, Albin Madapally Jose, @FartashFg , Josh Gardner & Chung-Cheng Chiu provides the answers! πŸ“° Read more: arxiv.org/abs/2510.20860

Data-Centric Lessons To Improve Speech-Language Pretraining

Spoken Question-Answering (SQA) is a core capability for useful and interactive artificial intelligence systems. Recently, several speech-language models (SpeechLMs) have been released with a...

arxiv.org
1
2
122
#PickOfTheWeek #SpeechLLM #LLM #MachineLearning #AI
1
1
95
MT Group at FBK retweeted
fbk_stek@fbk_stek
May 8
2
4
91
MT Group at FBK retweeted
Meetween@MeetweenEU
Mar 26
πŸŽ™οΈ Our paper on connecting Speech Foundation Models with LLMs is featured in the SpeechLMM Training Journal on Weights & Biases. Read it πŸ‘‰ bit.ly/4svG7ll SpeechLMM 2.0 coming this summer. πŸ‘€ #Meetween #SpeechLMM #AI #NLP
0:25
1
4
137
Load more