ASLP-lab

ASLP@NPU is a world-leading research group in speech, audio and language intelligence.

[ASLP-lab @AslpLab_NWPU]

🎙️ NVVSpeech Challenge — now open! #ISCSLP2026

Teach AI to hear & speak non-verbal sounds — laughter, sighs, coughs — in EN & CN.

🎯 T1: detect + transcribe NVV
🎯 T2: generate NVV speech

⏰ Submit by 7/29
🔗 https://t.co/vTECKaSSrZ

[ASLP-lab @AslpLab_NWPU]

Traditional speaker diarization pipelines: VAD → SV → ASR → error cascading 😵

SoulX-Transcriber says: one model, done. ✅

✅ Overlapping speech? 5-10 min meetings? No problem.
✅ Same-gender speakers? Distinguished.
✅ Output: timestamps + speaker labels + transcripts

[Nexdata.ai @nexdata_ai]

🚀 MLC-SLM Challenge 2026 Update: 𝗢𝗳𝗳𝗶𝗰𝗶𝗮𝗹 𝗕𝗮𝘀𝗲𝗹𝗶𝗻𝗲 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 𝗔𝗿𝗲 𝗡𝗼𝘄 𝗥𝗲𝗹𝗲𝗮𝘀𝗲𝗱

The 2nd Multilingual Conversational Speech Language Model (MLC-SLM) Challenge continues to advance research in 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝗺𝘂𝗹𝘁𝗶𝗹𝗶𝗻𝗴𝘂𝗮𝗹 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝘀𝗽𝗲𝗲𝗰𝗵, with a focus on speaker diarization, recognition, and deeper speech understanding.

This year’s training set has been further expanded to ~𝟮,𝟭𝟬𝟬 𝗵𝗼𝘂𝗿𝘀 𝗮𝗰𝗿𝗼𝘀𝘀 𝟭𝟰 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀, with richer language variants and accents to better reflect real conversational scenarios.

A key milestone is now here: 𝗧𝗵𝗲 𝗼𝗳𝗳𝗶𝗰𝗶𝗮𝗹 𝗯𝗮𝘀𝗲𝗹𝗶𝗻𝗲 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗳𝗼𝗿 𝗯𝗼𝘁𝗵 𝘁𝗮𝘀𝗸𝘀 𝗵𝗮𝘃𝗲 𝗯𝗲𝗲𝗻 𝗿𝗲𝗹𝗲𝗮𝘀𝗲𝗱.

For Task 1, the baseline is built on 𝗠𝗶𝗰𝗿𝗼𝘀𝗼𝗳𝘁’𝘀 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲 𝗩𝗶𝗯𝗲𝗩𝗼𝗶𝗰𝗲-𝗔𝗦𝗥, fine-tuned on the challenge training set, with standardized evaluation using tcpMER/tcpWER/tcpCER across languages.

For Task 2, the baseline combines 𝗚𝗲𝗺𝗶𝗻𝗶 𝟮.𝟱 𝗣𝗿𝗼 for constructing acoustic and semantic single-choice questions, and 𝗤𝘄𝗲𝗻𝟮.𝟱-𝗢𝗺𝗻𝗶-𝟳𝗕 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲𝗱 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲 𝗺𝘀-𝘀𝘄𝗶𝗳𝘁 𝘁𝗼𝗼𝗹𝗸𝗶𝘁. The development-set questions and answers will also be open-sourced for participants’ reference.

🏆 𝗧𝗼𝘁𝗮𝗹 𝗣𝗿𝗶𝘇𝗲 𝗣𝗼𝗼𝗹: 𝗨𝗦$𝟮𝟬,𝟬𝟬𝟬（ equivalent value in other currencies）
The prize breakdown for the top-performing teams in each task is as follows:
🥇 𝟭𝘀𝘁 𝗣𝗹𝗮𝗰𝗲: 𝗨𝗦$𝟱,𝟬𝟬𝟬
🥈 𝟮𝗻𝗱 𝗣𝗹𝗮𝗰𝗲: 𝗨𝗦$𝟯,𝟬𝟬𝟬
🥉 𝟯𝗿𝗱 𝗣𝗹𝗮𝗰𝗲: 𝗨𝗦$𝟮,𝟬𝟬𝟬

It’s exciting to see strong interest from both academia and industry, with teams already formed by leading global companies and renowned universities.

If you’re working on Speech LLMs, ASR, speaker diarization, spoken dialogue understanding, or multilingual AI, now is a great time to join.

Learn more and join us: https://t.co/LwaPWPwTFA
Register: https://t.co/6X1VlnRAzV

[ASLP-lab @AslpLab_NWPU]

🧵 Something huge is coming for AI Glasses! 👓
🏆We launch the first-ever Egocentric Speech Interaction Challenge at IEEE SLT 2026.
📊100+ hours 4-channel egocentric audio, two tracks: 1v1 dialogue & multi-party meeting understanding.
💰$7,000 prize pool.
https://t.co/QPuSBstT0e