- The pricing notice now actually reaches you. On quiet dictations it could be dismissed in the same instant it appeared, and people already on a Pro trial were shown it needlessly. Both fixed.
更新日志
Lispr 有什么新东西。
Lispr 发布很快 — 有时一天不止一次。下面列出抢先体验开放以来,每个版本的变化。
每次安装也都会通过 Sparkle 从 appcast feed 自动更新。
- Lispr Pro is here — and the app now tells you so. A one-time notice explains the free weekly minutes and what Pro adds, with a button to your account. Dictation, translation, your dictionary and history stay free.
- Minute counting no longer depends on the diagnostics switch. Plan limits are counted with a separate identifier that exists only for counting — it is not part of diagnostics and builds no profile of you.
- Limits now lead somewhere. When you reach a weekly-minutes notice, or formatting skips because of its monthly free limit, the notice carries a button straight to your account instead of just stating the fact.
- Click-through, locked down for good. The waveform pill is now structurally incapable of intercepting a click — not just while hidden, but at every moment of its life. The first pill after launch appears instantly again.
- Notices dismiss themselves. Occasional info cards (trial reminders, weekly-minutes heads-up) now fade after a few seconds instead of waiting for your next dictation, and a stale timer can no longer hide a fresh one.
- Click-through fix, part two. The waveform pill could also leave a click-blocking zone on newer macOS after dictation ended. Hidden overlays now leave the window list entirely and re-assert click-through on every show. If clicks near the bottom of your screen were dying — this ends it.
- Fixed an invisible window that could swallow clicks. On newer macOS, a hidden Lispr panel kept intercepting mouse clicks in a small zone near the bottom of the screen, making buttons of other apps unclickable until Lispr was restarted. Hidden panels are now fully click-through. Thanks to the user who sent a brilliant bug report.
- Founding invitations, better paced. The invite now reappears once a day while your window is open; dismissing it snoozes it for three days, claiming your place stops it for good.
- Founding 1000 invitations look the part. A refined glass banner with the Lispr mark, 30 seconds on screen — and a permanent «Founding 1000» menu entry while the window is open, so a missed banner never costs you the place.
- Fixed invisible in-app notices on newer macOS. A window-sizing bug could collapse banners (mic tips, update notices) to nothing; they now always render.
- In-app notices play nicer together. The update banner no longer covers other notices, and one-time messages wait for a quiet moment instead of disappearing unseen.
- Account groundwork continues. The app learned to show occasional account notices when they matter — nothing changes today.
- Your sign-in now stays signed in. Sessions renew quietly in the background, so the Account menu keeps showing your plan without asking you to sign in again.
- Fairer free minutes, gentler edges. Groundwork for weekly free minutes: if a cap ever applies to your account, you'll be warned in advance, your first time over comes with bonus minutes, and nothing you dictate is ever lost — recordings are kept and transcribed later.
- Clear feedback when a headset delivers no audio. If your microphone records pure silence (a common corporate-headset failure — often the headset's own mute button), Lispr now says so right away and offers a one-click switch to the built-in microphone, instead of silently doing nothing.
- After a silent recording from an external microphone, Lispr now re-opens the audio session automatically — many “dead mic” headsets recover without replugging.
- Input volume at zero or a muted microphone is now diagnosed on the very first attempt, with a one-click fix.
- Snappier start and finish for every dictation. Lispr no longer re-scans your audio devices on every keypress — the pill and start sound appear sooner, and your text lands faster after you release the key.
- Diagnostics now always describe the exact microphone that recorded each dictation — even if you switched devices mid-way.
- Instant dictation with AirPods and other Bluetooth headsets. Opening a Bluetooth microphone takes macOS 1–3 seconds — long enough to swallow your first words. Lispr now records from your Mac's built-in microphone instead while your headphones keep playing audio: dictation starts instantly, first words are never lost, accuracy is better, and your music no longer drops to phone-call quality. Prefer the headset mic? Toggle it off in the Microphone menu — an explicit mic choice is always respected.
- The pill now says “microphone…” while a mic is waking up. When the microphone genuinely needs a moment to start (Bluetooth headset with the lid closed, some USB mics), you'll see a clear connecting state instead of a frozen line — so you know exactly when your words are being captured.
- The Microphone menu now shows which device the automatic mode is using right now, and marks Bluetooth headsets that are slow to start.
- Much faster recovery on unstable connections. If the network silently drops during a dictation, Lispr now notices it right after you release the key and delivers your text in about 6 seconds instead of 20+. Nothing is lost — the audio is resent automatically.
- Your Lispr account — optional and passwordless. New “Account” item in the menu: sign in with a one-time email code, and your plan follows you across your devices. Google and Apple sign-in work on your account page on the web. Nothing changes if you skip it — dictation works without an account, as always.
- Your usage at a glance. The account page shows your dictation minutes and connected devices, with remote sign-out for any of them — open it straight from the menu.
- Tell us what’s wrong — right from the menu. New “Send Feedback…” item: report a problem or share an idea, optionally attaching technical diagnostics — app version, OS, device and microphone info plus the technical log. The text of your dictations is never included. Your message reaches the team instantly.
- “Check for Updates…” is now always in the menu. It used to appear only while holding ⌥ Option — many of you simply never found it. Manual update checks are now one click away.
- A dropped connection no longer stalls your dictation. When the network dies mid-dictation — hotel Wi-Fi, flaky tethering — Lispr now notices within seconds and immediately retries over a fresh path, so the text still arrives a few seconds later instead of a spinner hanging for minutes. Long dictations on slow uplinks are safer too: an upload that is still making progress is no longer cut off by a fixed timer.
- A stuck dictation can no longer freeze the app. If the connection dropped at the wrong moment, the pill could spin forever and every later key press was silently ignored — only restarting helped. Now such a dictation ends with a clear error, and the recorded audio stays in History so you can run it again.
- The assistant flip is snappier. The gradient now rolls across the wave about a second earlier — tighter pause detection and a quicker sweep.
- The pill now shows the assistant waking up — while you're still talking. Say «Лиспр, напиши…» and a couple of seconds in, a brand gradient rolls across the live waveform: understood, dictate the brief. Purely a visual hint — recognition and your text are unaffected, and it appears only for those who already use the copywriter.
- The copywriter's ear moved to the cloud. Recognizing the wake phrase («Лиспр, напиши…») now happens server-side — new spellings and fixes reach everyone instantly, without waiting for an app update. If the server is unreachable, dictations are always inserted verbatim — your text is never at risk.
- Setup Guide starts fresh. Reopening the guide from the menu now begins at the first screen — it used to stay stuck on the final one after you'd been through it once.
- Latin spellings of the name are matched by shape too. “Lispur” and similar renderings now wake the copywriter, completing the 0.89 change for Cyrillic — while real words like the insulin “lispro” stay untouched.
- Finding where first-run setup loses people. The app now reports anonymous setup progress — step names and timings only, no text and no audio — under the same “Diagnostics & statistics” switch. It's the groundwork for a fully redesigned onboarding.
- The copywriter now recognizes its name however Whisper spells it. Instead of a fixed list of spellings, Cyrillic renderings are matched by shape — «Лиспер», «Лиспор», «Лиспра» and future variants all wake the copywriter, while real words (including the Lisp programming language and its case forms) are carefully excluded.
- The copywriter answers to more pronunciations. Whisper sometimes hears the name as “Лиспер” or “Лиспар” — those now wake the copywriter too, caught in the first minutes of real-world testing.
- Compose straight into another language. Hold your translate key and say “Lispr, write to Ivan that we're moving the meeting” — you get the finished message in the target language, not a translation of your instruction. One step, ready to send.
- Fewer accidental compositions. Dictating about Lispr (“Lispr and other systems…”) no longer triggers the copywriter — the wake word now requires a pause or comma after the name, just like when you actually address it.
- History stays tidy when composing fails. A failed composition no longer leaves a phantom “recovered recording” duplicate after restart.
- Dictation no longer picks up numbers you never said. Lispr used to give the recognizer a small hint about how to write times. On longer dictations that hint could leak into your text as a run of stray numbers — and worse, push out words you actually spoke. The hint is gone. What comes back is what you said.
- A note on times. Because that hint is gone, a spoken time may now come back as words (“час тридцать”) rather than digits. We measured it: this affects about four dictations in a thousand, while the hint was costing more than it fixed. We would rather return your words faithfully and get the formatting right another way.
- Say what you want written, and Lispr writes it. Begin a dictation by calling Lispr by name — “Lispr, write to Ivan that we're moving the meeting to 3.30” — and instead of typing out your words, Lispr composes the message. Everything else is dictated exactly as before: the copywriter only runs when you call it.
- History keeps what you actually said. When Lispr changes your words — composing, formatting or translating — history now shows your original next to the result, so you can read or copy either one.
- No more stray numbers in long dictations. In Russian and Ukrainian, a long dictation could come back with a run of numbers stuck to it — “2. 3. 3. 4. 5.” — and, worse, missing pieces of what you said. Fixed.
- Dictated times come out as clean digits. When you spoke an appointment time in Russian or Ukrainian, Lispr could write one time as words and the next as digits in the same sentence — "час тридцать" next to "2.30" — leaving you to fix it by hand. Lispr now guides the recognizer to write times consistently as digits.
- 初始设置不会再卡在"输入监控"这一步。在某些 Mac 上，这一步可能会打开一个空白的设置面板，没有任何可以打开的开关，把你困在中途，无法完成设置——其实即便没有它，Lispr 的听写也能正常工作。现在这一步已改为可选，并会正确注册 Lispr，因此引导设置总能顺利完成。
- 麦克风现在能从更顽固的故障中恢复。在少数 Mac 上，麦克风可能会卡在出错状态，怎么重试都无法启动——现在 Lispr 会从头重建它的麦克风连接，把它重新唤醒，而不是让你对着一个失灵的麦克风干着急。
- 停顿之后，第一次听写不再慢吞吞。当 Lispr 闲置一段时间后，下一次听写可能会因为网络连接已经失效而卡顿好几秒。现在，Lispr 会在闲置一阵之后重新建立网络连接，让你即使中途休息过，听写依然保持流畅。
- 修复了打开历史记录时的卡死问题。在某些 Mac 上，打开历史记录窗口可能会让应用卡住、失去响应——Lispr 现在以更高效的方式渲染列表，因此历史记录可以即刻打开，并且全程流畅顺滑。
- 确认麦克风恢复功能确实有用。在后台，Lispr 现在会记录它的自动恢复功能在何时悄悄挽救了一次本会丢失的听写——这样我们就能看到 0.78 版本的修复正在为你发挥作用。你的听写方式不会有任何改变。
- 反应慢的麦克风不再让你白白丢掉一次听写。当麦克风需要多一点时间才能启动时，Lispr 现在会短暂等待第一段声音，而不是直接丢弃这次录音——这样那些很短或刚开口的听写也能顺利完成。它还会在你的 Mac 从睡眠中唤醒后重新预热麦克风，让第一次听写不再成为丢失的那一次。
- 启动不了的麦克风现在会自愈。修复了某些 Mac 上的情况:口述已经开始却没有采集到任何音频——Lispr 现在会检测到麦克风被中断或从未就绪,并在你下次录音时自动重新就绪。
- 粘贴现在能落到 Telegram 以及其他 Qt/Electron 应用里了。修复了一个问题:口述的文本进入了你的历史记录,却悄无声息地未能插入到聚焦的输入框——Lispr 现在会按下真正的 Command 键来粘贴,这样文本会出现在你打字的地方。
- 静音时自动停止。忘了结束口述?Lispr 现在会在一段安静之后自行停止。在设置中选择超时时长——关闭、30 秒、1 分钟或 2 分钟——现已本地化到全部 35 种语言。
- 更聪明的幻觉过滤器。一次安静或静音的轻按不会再变成凭空捏造的短语,例如「Thank you」或乱入的字幕署名——Lispr 现在会把转写结果与它实际听到你声音的位置对齐,并丢弃其余部分,任何语言都适用。真实的语音,哪怕是微弱的耳语,都会被保留。
- 你的音频能在崩溃中幸存。录音现在会边说边自动保存,在强制退出或崩溃后可从历史记录中恢复,长口述也不会再被截断——录音上限从 10 分钟提升到了 2 小时。
- 一个小型维护版本,旨在精进我们在现场对静音与语音的测量,好让即将到来的自动停止调优能恰到好处地落地。
- 对静音检测器的底层调优。为更聪明地判断「这是真实语音还是只是背景噪声」打基础——暂无行为变化。
- 幕后的可靠性和站点改进。口述行为没有变化。
- 修复了某些 Intel Mac 上的启动崩溃。0.70 的一处改动可能导致应用在某些 Intel 机型上启动时崩溃——本次更新已解决。如果你受到了影响,请从 lispr.ai 获取最新构建。
- 对设备端语音检测工作的内部调优。无面向用户的变化。
- 更聪明的静音检测,在设备本地完成。为完全在你的 Mac 上区分真实语音与安静/噪声打基础——这是消除静音片段有时凭空冒出的「Thank you」的根基。本版本仅做测量,尚不改变实际插入的内容。
- 失效的麦克风现在会告诉你。如果你的麦克风根本没真正开启,你会看到清晰的「麦克风未开启——请重试」,而不是一个悄无声息的空文本框——再也不会有无法解释的丢失口述了。
- 更克制的更新提醒。当某个版本修复了一个已知问题时,会有一条安静的应用内提示让你知晓——不会拿版本号给你难堪,也不会在口述途中弹窗。
- 加固的更新完整性。更新现在会在送达你之前于服务端验证,这样在传输途中损坏的下载会自动重试,而不是以「签名不正确」的错误失败。
- 更聪明的速度追踪。Lispr 现在能测量从松键到文本上屏的真实耗时,以便我们追查现实中的延迟——只是数字,在你的文本插入之后才发送,对口述速度毫无影响。会遵循你的诊断设置。
- 一个清晰的隐私开关。原来的「出错时发送」开关现在变成了「诊断与用量统计」,统一控制所有匿名技术数据——关闭它,任何技术数据都绝不会离开你的 Mac。
- 更简明的设置。两个数据控件现在并排放置,隐私政策也明确写清了每一项各自涵盖什么。
- 选择加入,帮助改进 Lispr。新增的一个设置开关让你可以分享录音,以便我们让口述更准确。它默认关闭,在你读清楚究竟会分享什么之后再由你决定。
- 未经你同意,什么都不会外传。这项贡献完全可选,且独立于正常口述——除非你打开它,否则你的音频和文本始终保持私密。
- 给所有人的全新构建。通过自动更新把近期的语言准确度和「精简」格式化修复推送给所有用户。
- 修复了俄语被变成乌克兰语的问题。0.60 引入的自动语言功能可能发生漂移,对部分用户强制套用错误的语言,改写了他们实际说出的内容——它现在已被禁用,任何受影响的配置都会在首次启动时重置。你仍然可以从菜单手动锁定一种语言。
- 「精简」格式化现在真的会压缩了。面对客气、铺垫很多的话语,「精简」模式此前会把一切都留成一整块——现在它能可靠地结构化并删减,同时保持你的语气和原意完好无损。
- 边说边整理你的口述。新的「格式化」菜单加入了两种可选模式——「轻度」会把你的话分成段落和列表,一个字都不漏;「精简」还会去掉口头禅和结巴。默认为关闭,除非你打开它,否则一切照旧。
- 更可靠的麦克风增益帮助。修复了一个问题:「调高输入音量」的提议从未真正向最需要它的人显示出来。
- 选择你的口述语言。新菜单让你锁定 Lispr 转写所用的语言——修复了乌克兰语和白俄罗斯语短语被误认成其他语言的问题。
- 它会学习你常用的语言。Lispr 会悄悄记住你用什么语言口述,并随时间逐渐倾向于那种语言,胶囊上还会显示一个源语言徽章。
- 轻声细语不再被丢弃。静音检测现在会查看你说话的峰值和信噪比,这样轻柔或远处的语音会被转写,而不是被当作「无音频」扔掉。
- 一键修复麦克风。当你的输入音量太低听不清时,Lispr 现在会提议帮你调高——并尽早发现失效的 AirPods 或外接麦克风,免得一段长录音白费。
- 更少的误报「未检测到音频」横幅——偶发的一次小故障会保持安静,直接重试,而不是反复打扰你。
- 随时随地打开历史记录。在任意应用中按
⌃⌘Y即可调出你的口述历史,无需切回 Lispr。 - 修复了 macOS 12 Monterey 上的空录音。一个编解码器的怪癖会在 Monterey 上悄悄截断音频,导致口述返回空白——现在它们能正常转写了。
- 对俄语和韩语措辞的本地化打磨。
- 打基础的版本。主要是底层工作,用来追查一个罕见的「录音过短」失败,外加本地化打磨。对大多数用户没有行为变化。
- 麦克风助手现在能诊断更多情况。它可以识别出毫无信号的失效麦克风、被调低的输入音量滑块、把你音频压平的语音处理软件,或是嘈杂到无法转写的环境——每种情况都配有针对性的建议或一键跳转到对应设置。
- 麦克风助手的所有提示文案现在都已人工翻译到 Lispr 的全部 35 种语言。
- Lispr 现在会告诉你它为什么听不到你。连续几次录到静音后,胶囊下方会出现一条横幅,说明可能的原因——并提供一键修复。
- 如果你选的麦克风已断开(AirPods 取下、USB 拔掉),横幅会提议帮你切回内置麦克风。如果麦克风访问权限被关闭,它会直接带你到对应的设置面板。
- 更精准的静音录音诊断——Lispr 现在能识别出因合盖处于贝壳模式而导致内置麦克风被捂住的情况。同时修复了一个影响 macOS 11 目标的构建问题。
- 翻译会从你实际所说的内容中学习。当你边口述边翻译时,Lispr 现在会从原始语音(而不是翻译后的结果)中提取你的自定义词汇——这样目标语言里的日常词就不会再被误当成你的品牌词。
- 历史记录会保留原文。翻译过的口述会显示译文并带一个小小的
→CODE徽章,而你说出的原文则保留在下方。
- 实时用量计数器不再抽风。lispr.ai 上计数器背后的统计接口大约每三次加载就有一次返回 504——现在它直接与
proxy.lispr.ai通信,加载稳定可靠。
- 麦克风故障可见化。当你明明在说话、录音却返回空白或静音时,Lispr 现在会记录麦克风、音量和权限状态,这样就能查出失效或选错的输入设备,而不是悄无声息地失败。
- 即时翻译——说一种语言，打另一种。 像往常一样口述，任何时候在句子中途按下第二个键：Lispr 会插入翻译而不是原文。按住 ⌃ 选择一种目标语言，按住 ⇧ 选择另一种。录音胶囊会显示一个小 →EN 徽章，让你随时知道会得到什么。提前松开触摸仍然计入。
- 两个可配置槽位，支持32种语言。 在新建的 Translate 窗口中为每个槽位选择键和目标语言。首次启动时会根据您的 Mac 语言自动选择合理的默认设置。如果翻译失败，Lispr 会回退插入纯文本转录——您永远不会失去所说内容。
- Lispr 自动学会你的品牌名。全新的「词汇表」功能会观察 Whisper 容易把你常用的词听错成什么——把 Codebridge 听成「Котбридж」、把 Monobank 听成「Монобанк」——然后悄悄把正确的拼写反馈给模型。默认启用,这样你无需四处寻找隐藏开关就能获得准确度提升。从菜单打开「词汇表」即可查看学到了什么、编辑任意条目或将其关闭。
- 菜单做了一次大幅精简。顶层条目从 17 项减到 9 项。触发键和触发侧合并进一个「触发」子菜单。不常用的项目(登录时启动、诊断、设置向导、辅助功能)移进单个「设置」子菜单。「关于」挪到了底部,以贴合 macOS 菜单栏应用真正的使用方式。「检查更新」现在成了「关于」的 Option-alt 项——按住 ⌥ 即可显示。
- 选择触发键的哪一侧生效。如果你用 option+e 在 PT/FR/ES/DE/IT 中输入重音符号,现在可以把口述限制为只用右 Option,把左 Option 留给排版输入。每个键各自记忆——Option 的偏好不会影响 Shift 的。可在引导流程中以及菜单的「触发」下设置。
- 菜单提示会跟随你的触发键。如果你选了 Control,提示现在会显示「按住 ⌃ 说话」——而不是以前无论你怎么选都写死的、会误导人的 ⌥ 字形。引导文案和触发键名称在全部 35 种语言中都是如此。
- 历史记录和词汇表现在感觉像同一个应用。共享的外观、字体和间距。两者都使用系统窗口背景——这意味着历史记录终于能在深色模式下正常显示了(它写死成米色背景已经有六个月之久)。
- Whisper 提示不再偏向俄语。词汇此前是以
"Имена: Codebridge, Monobank."这样带西里尔字母前缀词的形式送进 Whisper 的。在英语/葡萄牙语/印尼语音频上,那个词会把 Whisper 的语言分类器拉向西里尔字母输出。现在它是一个纯粹的逗号列表——对专有名词同样的准确度提升,却没有语言偏向。 - 词汇表界面变得更清爽。移除了彩色的「置信度圆点」和计数列——它们并不会促成任何用户操作。编辑和删除现在被放到了显眼的位置。
- 词汇表功能的自动诊断。当词汇表对真实用户出问题时(文件损坏、保存失败、连续三次重试都连不上服务器),我们现在能及时知道——这样问题在扩散之前就能被修复。
- Lispr 学会你的品牌名。全新的「词汇表」功能会观察 Whisper 容易把你常用的词听错成什么——把 Codebridge 听成「Котбридж」、把 Monobank 听成「Монобанк」——然后悄悄把正确的拼写反馈给模型。从菜单打开「词汇表」即可查看学到了什么、编辑任意条目或删除某个词。
- 菜单做了一次大幅精简。顶层条目从 17 项减到 9 项。触发键和触发侧合并进一个「触发」子菜单,不常用的项目移进单个「设置」子菜单,「关于」挪到了底部,以贴合 macOS 菜单栏应用真正的使用方式。
- 选择触发键的哪一侧生效,提示也会随你的选择变化。可以把口述限制为只用右 Option,把左 Option 留给在 PT/FR/ES/DE/IT 中输入重音符号——每个键各自记忆。菜单提示现在会按你的实际触发键显示「按住 ⌃ 说话」,而不是写死的 ⌥,所有 35 种语言都是如此。历史记录也用上了正确的深色模式。
- 针对 0.44 上线的后续修复——解决了一个可能让下载页面短暂无法访问的小问题。另外还加入了更安静的底层时序埋点,用来追查最后那几例罕见的口述变慢报告。
- 网络不稳时录音胶囊不再卡死。现在有 30 秒看门狗,一旦上传卡住就直接取消,而不是让胶囊永远悬在那里——罪魁祸首是时断时续的移动 Wi-Fi 上残留的连接池连接。Lispr 还会在你的 Mac 唤醒的瞬间刷新连接,这样睡眠后的第一次口述就不会卡住。
- 更聪明地就近接入边缘节点。Lispr 现在按真正处理你流量的数据中心来选服务器,而不是按你的账户所在国家——这修复了部分用户在酒店和强制门户 Wi-Fi 上遇到的莫名其妙绕道悉尼的问题。
- Lispr re-warms its connection to the proxy as soon as your Mac wakes up — the first morning dictation no longer pays a TLS handshake.
- Connection reuse across dictations. One shared TLS connection survives every dictation in a session — eliminates the 2–3 second lags some users hit on flaky Wi-Fi when each upload had to re-handshake.
- CRC self-test on the Opus framer expanded from 1 to 5 vectors — defends against a class of bug that previously crashed 0.41 on its first launch.
- Opus 流式上传 + 地理感知中继。从松键到文本的中位数降到 ~300 ms(此前多数流程是 1.5–2.5 秒)。
- lispr.ai 上的未知 URL 现在会返回真正的 404,而不是悄悄端出首页。
- 诊断用
X-Lispr-*响应头 — 不在服务端存日志也能远程查看 Lispr。
- 短的和空的录音现在会在送去转写之前过滤掉 — 解决 10 条里大约 9 条假报的"没听清"。
- DMG 和 Sparkle appcast 走 TCP-only 代理(
proxy.lispr.ai)— 修复了一些严格网络下的自动更新。
- Esc 取消录音。插入之前随时按一下,就能把刚说的话扔掉。
- Whisper "幻觉"过滤 — 安静的录音不会再变成编造的句子。
- 录音长度软上限(10 分钟)带本地化提示 — 避免不小心录马拉松。
- 亚太流量走悉尼中继 — 对该地区比走美国路径明显更快。
- 默认直连 Groq;被地理屏蔽的路线退到美国的 Durable Object 上。
- TCP-only 区上的自定义代理域名
proxy.lispr.ai— 绕开一些网络的 HTTP/3 问题。 - 移动落地页打磨:页脚内边距、主按钮对齐、语言菜单关闭行为。
- 继续口述时不再多一个前导空格 — Lispr 只会在自己上次插入之后才自动拼接。
- 转写超时从 30 秒砍到 10 秒 — 网络抖一下恢复得更快。
- Lispr 现在能正确注册到 系统设置 → 隐私与安全 → 输入监控(通过
IOHIDRequestAccess)— 入门流程更干净。 - Groq 密钥智能轮换,密钥耗尽时还会提醒 — 不再因配额用完而悄悄失败。
- App 自带
libswift_Concurrency.dylib— Lispr 在 macOS 11 Big Sur 上能干净地启动。 - 下载 URL 现在带
Cache-Control: no-store— 重新下载总会拿到最新签名构建。
- 短暂转写失败时自动重试。
- 可以从历史视图重试一次失败的口述。
- 智能拼接句子。连续的口述之间自动加上句号和空格 — 不再有甩不完的长句。
- 菜单栏的录音胶囊现在在各阶段(在听 · 在转写 · 完成)之间会平滑伸缩,不再生硬切换。
- Groq 400 错误时不再无声失败,而是友好提示"没听清 — 再来一次"。
- 落地页的实时计数器真的实时更新了(以前只是快照)。
- 节省时间的计数器现在沿整条单位阶梯一路爬:分钟 → 小时 → 天 → 年 → … → 千年。
- 双击锁定录音。按住一下来口述,或者双击来锁定 — Lispr 会一直免提录下去,直到你再点一下。
- Fn 键作为按下说话的触发。用 Fn 代替(或并用)右 ⌥ — 如果 Option 已经被占就很方便。
- 菜单栏图标在 macOS 11–12 上不再消失。
- 入门提示中的火花改成用形状绘制 — 不再依赖旧版 macOS 缺少的 SF Symbol "sparkle"。
- 最低 macOS 从 14 下调到 11。Lispr 现在能跑在所有 Big Sur 及之后的版本 — Apple Silicon 和 Intel 全覆盖。
0.2 – 0.23 · 抢先体验上线
17–18 May 2026
最初 22 个版本 — 首次公开发布。通过右 ⌥ 实现按下说话,通过 Groq 在边缘网络上调用 Whisper 进行转写,~4 MB 的菜单栏 App,自动识别 ~99 种语言,麦克风选择器,经 notarize 的 DMG 配 Sparkle 自动更新,以及落地页上的实时使用计数器。