「锦秋AI实验室」 而谁还停留在"懂了点，又好像没懂透，反正先瞎操作一波"的阶段？ 这是一档专注于探索和评测AI产品在实际场景中应用效果的栏目。 我们正在 用AI 解锁100个效率场景。 下一个场景会是什么？ 以前以为"语音输入"只是给懒人用的：张嘴说两句，手机替你打字。 直到真的开始用它写长文、回微信、记灵感、开会做纪要——才发现，语音输入法的核心根本不是"省事"，而是： 它到底能不能把我说的"人话"，变成 能能让信息接收者听懂 的 "人话"。 我们也被这些"转写翻车"折磨过。 所以这次我们决定认真测一测： 7 款 AI语音输入法，5个真实场景，统一题库，一轮一轮地比。 我们想知道： 在语音输入这件事上，谁真的听懂了"帮我把我说的话打出来"？ * 需要说明的是 ，我们此系列的测评以年轻普通用户的实用视角和审美进行测评，于 AI 产品持有相对积极的评价态度。 这里也插播一下未来的测评预告： 近期我们还将会进行 AI 小游戏制作、 AI 知识库、 AI 画布、 AI 陪伴类产品的测评。如果你对这些 AI 产品方向的测评感兴趣，也欢迎私信或者 评论区告诉锦秋基金（微信公号：锦秋集；微信 ID ： jqcapita ...

真正的转折，其实发生在我开始高频使用各种 AI App 的这两年。 第一次真正觉得「语音输入这件事好像值得重视」，是各个 AI App 里那个「语音转文字」按钮变得越来越好用的时候。这些 App 里的语音转写，明显比 传统输入法里的语音要聪明得多：它不仅能听清我在说什么，还能自动加上标点，帮我把一些口语化的表达整理得比较书面，甚至在我说得磕磕绊绊的时 候，最后呈现出来的那一段文字读起来仍然是顺的。 如果几年前有人跟我说，「你以后写稿可能不怎么需要键盘了」，我大概会把这句话当成一句玩笑。那时候我正处在对机械键盘的迷恋期，研究轴体、键 帽、键程，购入过 Cherry、Filco、NiZ、Keychron、3D 打印分体式键盘。甚至为了提高打字效率，专门学习过双拼输入法。 我的注意力都放在消费的快感上，很少认真想过这样一个问题： 敲键盘，真的是输入的最优解吗？ 主流的 AI 几乎都覆盖了语音转文字功能｜图片来源：极客公园 更关键的是，它和后面的 AI 是连在一起的——我说完一句话，看到的不只是干巴巴的转写结果，而是 AI 根据这段话给我的反馈和回答。那一刻我第一 次有了一个直观的感受：语音不再只是一个「替代键盘的输 ...

Core Insights - The recent surge in large models has unexpectedly revitalized the input method sector, previously considered a basic infrastructure, making it attractive by the second half of 2025 [1]. Group 1: Market Developments - In the past two months, there has been a significant increase in news density regarding voice input technologies, with major developments from both domestic and international players [2]. - Domestic advancements include ByteDance's Doubao input method officially launching after internal testing, and WeChat input method continuously iterating on AI-assisted features [2]. - Internationally, Wispr announced a $25 million Series A funding round, bringing its total funding to $81 million, while Typeless gained attention on Product Hunt [2]. Group 2: Competitive Landscape - The voice input market can be categorized into three main camps: 1. Desktop SaaS players like Wispr and Typeless, focusing on productivity for core office users. 2. Mobile giants like Doubao and WeChat, leveraging vast ecosystem traffic for social interactions. 3. Low-cost indie developers represented by Whisper Keyboard and Lightning Say, focusing on localized or independent development [4]. Group 3: Product Performance - A subjective testing scenario revealed Typeless as the best desktop input method and Doubao as the best mobile input method, with specific strengths in handling complex language and context [6]. - Typeless achieved a processing time of 3.05 seconds, effectively removing filler words and correcting formats, while Doubao excelled with a 2.05-second response time, accurately interpreting context [6][13]. - WeChat input method, with a rapid 1.08 seconds response time, remains dominant in casual communication despite some limitations in professional formatting [13]. Group 4: User Experience Insights - The user experience of third-party voice input methods on iOS is often hindered by permission issues, requiring app switches for voice input [8]. - Doubao's voice model demonstrates superior performance in speed and accuracy, particularly in Chinese, although it faces challenges on iOS due to Apple's privacy restrictions [8][42]. - Typeless offers the best output quality for desktop users, providing high accuracy and innovative interaction features, while Lightning Say, despite its speed, struggles with professional terminology [8][60]. Group 5: Technological Evolution - The voice input sector is experiencing a paradigm shift from traditional automatic speech recognition (ASR) to models that understand and reconstruct language, enhancing user interaction [63]. - This evolution allows for greater tolerance of user errors, enabling a more natural and intuitive communication interface, transforming input methods into tools for thought rather than mere transcription [64][65].