Files
icassp2027-experiment/literature_notes/36_flora_device_directed.md
2026-08-30 15:36:28 +08:00

14 lines
1.0 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Multimodal Large Language Models with Fusion Low Rank Adaptation for Device Directed Speech Detection
- 第一作者:Shruti Palaskar;Apple
- 发表:Interspeech 2024
- 阅读深度:重点方法、输入与指标章节;未运行
- 记录日期:2026-08-28
- 来源:[ISCA 论文](https://www.isca-archive.org/interspeech_2024/palaskar24_interspeech.pdf)、[Apple](https://machinelearning.apple.com/research/llm-fusion-low-rank)
## 摘要与项目关系
FLoRA 通过模态投影、低秩适配及缺失模态处理,让预训练语言模型吸收音视频,判断语音是否对设备发出。使用 BART/T5 系列与私有数据,采用 EER 和固定 false reject rate 下的 false accept rate。
两点可借鉴:固定召回比较误触发有先例;局部特征可以通过训练适配进入模型,不必局限最终拼接。其任务是 utterance-level device-directed detection,不是 wearer onset。当前均值向量虽可训练映射,却不能被声称保留了原生细粒度媒体 token 的全部信息。