Language Model Estimations and Representations for Real-Time Continuous Speech Recognition

25 Nov

🚀 JOB ALERT: The FBK's MT Unit is hiring!

Join us as a Researcher in Multimodal LLMs and work at the forefront of AI, combining text, speech, and video. If you’re passionate about deep learning and next-gen models, we want to hear from you!

Apply 👉 https://jobs.fbk.eu/Annunci/Jobs_A_Researcher_in_Multimodal_Large_Language_Models_LLMs_241756954.htm

Reply on Twitter 1993370193146569123 Retweet on Twitter 1993370193146569123 7 Like on Twitter 1993370193146569123 4 Twitter 1993370193146569123

Retweet on Twitter MT Group at FBK Retweeted

Avatar BeatriceSavoldi @beatricesavoldi ·

25 Nov

🚀 We're hiring a Researcher in Responsible & Trustworthy NLP!
Join our @fbk_mt group at @FBK_research to work on fairness and trustworthiness in multilingual technologies.
📅 Deadline: Dec 10, 2025
🔗 Apply: https://jobs.fbk.eu/Annunci/Offerte_di_lavoro_A_Researcher_in_Responsible_and_Trustworthy_NLP_241757983.htm

Reply on Twitter 1993244192836141091 Retweet on Twitter 1993244192836141091 8 Like on Twitter 1993244192836141091 5 Twitter 1993244192836141091

Avatar MT Group at FBK @fbk_mt ·

14 Nov

Heading home tired but very happy after a fantastic #EMNLP2025 (and some well-deserved vacation 😎).
We were super busy presenting 5 papers! It was fantastic catching up with colleagues, exchanging ideas, and seeing all the amazing work in the #NLProc community!

(1/6)

Reply on Twitter 1989399161842930115 Retweet on Twitter 1989399161842930115 2 Like on Twitter 1989399161842930115 8 Twitter 1989399161842930115

Avatar MT Group at FBK @fbk_mt ·

12 Nov

Our pick of the week by @FBKZhihangXie: "#Speech Discrete Tokens or Continuous Features? A Comparative Analysis for Spoken Language Understanding in #SpeechLLMs" by @WangDingdo2603, Junan Li, @HelenMeng_CUHK, et al. (#EMNLP2025)

#SLU #SpeechTech

Zhihang Xie @FBKZhihangXie

🚀 New paper: Speech Discrete Tokens or Continuous Features?
📄 https://aclanthology.org/2025.emnlp-main.1266.pdf
🧩 A comprehensive benchmark of SpeechLLMs using HuBERT/WavLM with Qwen & LLaMA.
✨ Continuous features outperform overall, while discrete tokens excel at phoneme-level detail.

Reply on Twitter 1988617563878699388 Retweet on Twitter 1988617563878699388 Like on Twitter 1988617563878699388 1 Twitter 1988617563878699388