📡 Idea Radar 每天 8:00 / 18:00 推送 · 18:00 生成精选
reddit · SideProject · 51 分 ← 返回列表

I built a local-first dictation app in Rust and Tauri, and I'd like a technical reality check please ;O)

I've been building NV oOS Cognition , a desktop dictation app. Hold a hotkey, speak, release: the speech recognition and the text cleanup run on your own machine, and the cleaned text pastes into whatever has focus. I'm the author, this is my project, and I'm posting because I want the architecture picked apart. Why another one Most of the good dictation tools are cloud services. You talk, the audio goes somewhere, text comes back, and you pay monthly. I wanted the boring version: my v

五维评分

付费意愿
55
需求真实度
45
个人可落地
60
竞争空间
30
新鲜度
55

AI 分析

痛点:主流听写工具都是云端订阅服务,音频要上传、按月付费,用户想要隐私可控、离线可用、一次下载模型就能本地推理的听写体验。

目标用户:注重隐私的开发者/写作者/知识工作者,尤其是经常口述代码、术语、需要自定义词表的重度键盘用户

证据:Most of the good dictation tools are cloud services. You talk, the audio goes somewhere, text comes back, and you pay monthly. I wanted the boring version: my voice stays on my machine, it works offline, and the model is a file I downloaded once.

变现:一次性买断的桌面 License(含一年更新),或免费核心 + 付费高级功能(自定义词表、per-app 配置、LLM 后处理、团队同步配置);也可做企业内网私有部署授权。

MVP:两周版:Tauri 桌面壳 + whisper.cpp/sherpa-onnx 本地 STT + 按住热键说话、松手粘贴 + 确定性文本清理(标点、大小写、数字规范化)。先只做 macOS 一个平台、一个模型,把「按下到出字」的延迟压到 1 秒内作为唯一卖点。

竞品:Wispr Flow、Superwhisper、MacWhisper、Aqua Voice、VoiceInk、Talon Voice

摘要:作者用 Rust + Tauri v2 自建了本地优先的桌面听写应用 NV oOS Cognition,支持 Parakeet TDT、Whisper、Moonshine 三种本地 STT,带预录环形缓冲、确定性文本清理和可选 LLM 润色,发帖只为求架构批评。帖中对比了云端听写服务「要上传音频、按月付费」的痛点,但没有任何付费意愿原话或收入数据,需求验证仅停留在作者个人主张。赛道已有 Wispr Flow、Superwhisper 等一批成熟付费竞品,竞争空间偏小。