Just talk, no script
Record a quick take, or import a vlog or talking-head video you've already shot. No script, no waiting until you're "ready".
Turn the foreign language you speak into subtitles, a study blog post and word-highlighted subtitle videos. Even if you stumble, SpeakLog transcribes it, puts what you said next to a more natural version, and helps you fix it one sentence at a time.
Transcribes, renders and exports on your Mac · Japanese, English, Chinese, Korean and more
Yesterday I went to a new café with a friend.
Many people wait until they're ready and can say it right, so they never say it at all. Those who do speak are left with nothing afterwards: no idea what went wrong, and nothing to show for it.
Record a quick take, or import a vlog or talking-head video you've already shot. No script, no waiting until you're "ready".
Raw recognition is retained separately from reviewed and natural wording. Compare the source recording with your practice record to revisit your expression and later edits.
Your original plus the improved version is a genuine study record. Export it as subtitles, a blog post or a video for the platforms you already use.
Drag in a voice memo, vlog or talking-head video, or hit record in the app. Imported files are referenced at their existing location.
On-device speech recognition with per-word timestamps and confidence, split into sentences at pauses. Speed depends on hardware, language and media.
AI flags misheard words and unnatural phrasing, with reasons. Accept, skip or edit each sentence from the keyboard.
Export subtitles, a study blog post and a subtitle video — in any order, or just the one you need.
昨日は友達と新しいカフェを行きました。
昨日、友達と新しいカフェに行きました。
ケーキがすごいおいしいでした。
ケーキがすごくおいしかったです。
SpeakLog only exports — there's no SpeakLog social network. Where your work goes is up to you.
Timed, reviewed subtitles. Keep your original words or use the improved version, and add a translation for bilingual captions.
CapCut · Aegisub · Final Cut · YouTube
A study note: your text first, then "what I said → more natural" pairs, grammar points and a vocabulary list. Templates keep a series consistent.
note · Medium · X · Obsidian · static blogs
Karaoke-style word highlighting, part-of-speech colors, readings and translations. Preview and export share one renderer, so what you see is what you get.
YouTube · Instagram · TikTok · Bilibili
English, Japanese, Chinese, Korean, French, Spanish, German… set per project.
Furigana for Japanese, pinyin for Chinese, romanization for Korean and Russian — or just part-of-speech colors.
Nothing to say? Draw a topic, or let AI suggest a batch matched to your level.
Hit record to start; stop and it becomes a project.
Tell it in one sentence how to fix subtitles, rewrite the blog or organize notes. Every change can be undone.
Detects faces on your Mac and lets you pick who to blur. You, the person filming, stay unblurred by default.
Import several videos and photos at once; AI writes an overview, rates each clip and analyzes content and visuals.
A built-in MCP server lets Claude Code or Codex read and edit projects and export videos.
Have AI write a lesson in the SpeakLog format; it builds Q&A cards, shadowing drills or vertical video diaries with captions. Record each line and export.
Play a model reading of any sentence before you speak — macOS voices, VOICEVOX, AivisSpeech or Azure. Unrecorded lines can use the voice for now.
Japanese sentences show pitch accent, phrase breaks and pauses, so you can read closer to natural speech.
AI groups corrections across recordings into recurring issues, with causes, fixes, practice sentences and trends.
Transcription, video rendering, face blurring and export run on the Mac. Cloud review, blogs and assistant features send the text and context they need; visual analysis passes sampled frames or photos to Claude Code or Codex. Selecting an on-device model does not make every feature offline. Read about data flow and AI costs
Installation, language and AI settings, and your first import.
Separate recognition errors from expression problems, then say it again.
Choose SRT / VTT / ASS, Markdown / HTML, or an MP4 video.
See which tasks run locally and which text or images go to AI.
Practice languages are set per project, including English, Japanese, Chinese, Korean, French, Spanish and German. Transcription depends on macOS speech support and may need a model download on first use. Explanation and translation languages are set separately.
Transcription, rendering, face blurring and export run locally. Cloud AI sends the text and context each feature needs. Visual analysis passes sampled frames or photos to Claude Code or Codex; it still uses Codex when the global provider is the on-device model. Check automatic review and provider settings before using private content.
Transcription and natural wording are separate layers, and review lets you accept, skip or edit suggestions. The AI assistant and external agents can edit subtitles and other project content, so separate layers do not make every field immutable. Keep the original media and review important changes.
The current 0.3.0 package requires an Apple silicon Mac (M-series chip) with macOS 26 or later. Intel Macs are not supported by this package. The on-device text AI also requires available Apple Intelligence.
Export SRT, VTT or ASS subtitles; save Markdown or HTML blogs, copy rich text to note or Medium, or split text into an X thread. Video exports are MP4 in 16:9, 9:16 or 1:1. You upload or publish the results yourself.
It doesn't have to be good — speak first, polish it later.
Download for MacCurrent version 0.3.0 · Apple silicon (M-series) · macOS 26+ · Apple-notarized installer · DMG and previous versions