PulseAugur
中
实时 06:29:59

Stride-k subsampling 将 Whisper 音频令牌减少 75%，无需重新训练

作者 PulseAugur 编辑部 · [1 个来源] ·

研究人员开发了一种名为 stride-k subsampling 的新方法，可以在无需额外训练的情况下减少 OpenAI 的 Whisper 模型处理的音频令牌数量。该技术涉及选择每 k 个令牌，并且发现 stride-2 配置可将音频令牌减少 75%，GFLOPs 减少 50% 以上，同时对大多数 ASR 基准测试的词错误率 (WER) 影响最小。该方法还显示出对基于 Whisper 的 SpeechLMs 的好处，将延迟最多减少 27.4%，而准确性仅略有下降。 AI

影响 降低音频处理模型的计算要求，可能降低推理成本和延迟。

排序理由 关于现有模型新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

Stride-k subsampling 将 Whisper 音频令牌减少 75%，无需重新训练

本文如何被排名

Signal score
30 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
关于现有模型新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, infra
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.AI TIER_1 English(EN) · Chanhee Cho, Junhyuk Choi, Bugeun Kim ·

    Stride-k Subsampling: Train-Free Audio Token Reduction for Whisper

    arXiv:2608.30927v1 Announce Type: cross Abstract: Whisper exposes speech through a fixed 1500-token encoder interface, now a default representation for ASR decoders and Whisper-based speech language models (SpeechLMs), yet its redundancy remains largely unexamined. We propose str…

相关实体

相关话题