恭喜實驗室 3 位成員從荷蘭 Interspeech 2025 發表完 papers 回國拉！附上 poster 版版照和大家分享，一起來看這次 BIIC Lab 的三篇研究吧！

Ⅰ｜Defend for Self-Vocoding: A Novel Enhanced Decoder Network for Watermark Recovery｜Yu-Sheng Lin
提出一個全新的「浮水印解碼架構」，即使聲音被重新編碼，也能完整找回隱藏的浮水印 
➧ 為保護聲音、對抗 AI 偽造攻擊跨出重要一步！

Ⅱ｜ZSDEVC: Zero-Shot Diffusion-based Emotional Voice Conversion with Disentangled Mechanism｜Hsin-Hang Chou
透過擴散模型 + 資訊解耦 + 情緒引導，我們能在「沒見過的講者」身上完成自然又精準的情緒語音轉換
➧ 未來虛擬助手、遊戲角色、AI 廣播員都能更生動！

Ⅲ｜Lessons Learnt: Revisit Key Training Strategies for Effective Speech Emotion Recognition in the Wild｜Jing-Tong Tzeng
重新檢視「被忽略的訓練細節」——微調模態、融合方法、focal loss 
➧ 成功讓語音情緒辨識在真實世界裡表現更穩、更準！

Congrats to our 3 BIIC Lab members for presenting their papers at Interspeech 2025 in the Netherlands and bringing back these lovely poster shots! Here's a quick look at the three works we showcased this year:

Ⅰ｜Defend for Self-Vocoding: A Novel Enhanced Decoder Network for Watermark Recovery｜Yu-Sheng Lin, Ching-Yu Yang, Hsin-Hang Chou, Ya-Tse Wu, Bo-Hao Su, *Chi-Chun Lee
A new watermark recovery framework that protects voices against self-vocoding attacks, keeping audio safe from AI forgery.

Ⅱ｜ZSDEVC: Zero-Shot Diffusion-based Emotional Voice Conversion with Disentangled Mechanism｜Hsin-Hang Chou, Yun-Shao Lin, Ching-Chin Sung, Yu Tsao, *Chi-Chun Lee
A diffusion-based emotional voice conversion model that works even on unseen speakers—making voice assistants, game characters, and AI announcers more expressive.

Ⅲ｜Lessons Learnt: Revisit Key Training Strategies for Effective Speech Emotion Recognition in the Wild｜Jing-Tong Tzeng, Bo-Hao Su, Ya-Tse Wu, Hsing-Hang Chou, *Chi-Chun Lee
Revisiting overlooked training details—fine-tuning, fusion, focal loss—to achieve state-of-the-art performance in real-world speech emotion recognition.

Big cheers to Yu-Sheng, Hsin-Hang, and Jing-Tong! Can't wait for BIIC Lab's next stage!!