New paper "MultiTalk" extends the Moshi paradigm to long, multi-party, bilingual conversations, releasing 57.6k hours of speech data and codec-frame-level full-duplex modeling for English and Chinese. Aims to address…

#OpenSourceAI #SpeechModels #Moshi #FullDuplex
https://arxiv.org/abs/2609.36903

🔥 Alibaba's Qwen drops Qwen-Audio-3.1! New models for speech recognition, synthesis, interaction & *creation*! Plus, major price cuts across the board: TTS down 70%, Realtime 85%, and ASR a whopping 95%! Get ready for powerful, affordable audio AI. #QwenAudio #AI #SpeechModels

Researchers found visual grounding boosts alignment between spoken and written representations, mainly via stronger encoding of word identity. Speech‑based encoders stayed primarily phonetic. https://getnews.me/visual-grounding-impacts-speech-and-text-language-models/ #visualgrounding #speechmodels

in the year of our lord 2026, Apple is still forced to use caseless-enums as namespaces in their own public frameworks

since Swift doesn't apparently have anything better

#wwdc

developer.apple.com/documentatio...