Files
icassp2027-experiment/literature_notes/37_whisper_flamingo.md
2026-08-30 15:36:28 +08:00

14 lines
1.1 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.
# Whisper-Flamingo: Integrating Visual Features into Whisper for Audio-Visual Speech Recognition and Translation
- 第一作者:Andrew Rouditchenko
- 发表:Interspeech 2024
- 阅读深度:重点方法、融合消融与官方仓库说明;未运行
- 记录日期:2026-08-28
- 来源:[ISCA](https://www.isca-archive.org/interspeech_2024/rouditchenko24_interspeech.html)、[论文](https://arxiv.org/html/2406.10082)、[代码](https://github.com/roudimit/whisper-flamingo)
## 摘要与项目关系
将 AV-HuBERT 视觉特征通过门控 cross-attention 融入 Whisper decoder,在噪声条件改善语音识别和翻译。门控从零开始,初始新增层为恒等变换;适配阶段保留强音频模型。
这是“内部学习视觉增量”的直接结构参考,比直接修改冻结 hidden state 更有训练依据。但原版不是流式接话模型,也不是自动不退化保证;仍需按可用前缀限制视频/音频。我们优先试少量上层或临时 readout query 的 adapter,不照搬每层宽 cross-attention 的参数开销。