Updated September 6, 2026 · By Sumbat.T

Obsidian Voice to Text: What Actually Dictates in 2026

There is a specific reason this question is harder to answer than it looks, and it is not that the information is missing. It is that the word “voice to text” covers two different jobs, most of the available advice quietly answers the wrong one, and the evidence people cite about what works has been passed around for so long that it no longer says what it is used to say. This page separates the two jobs, reports what each of the recommended plugins actually does according to its own documentation, and is honest about the one question nobody has actually settled.

The distinction almost every guide skips

Start here, because getting this wrong is what makes people install three plugins and conclude that dictation in Obsidian is bad.

Both are legitimate. They are simply different tools, and the search results for this topic do not distinguish between them at all. Someone who wants to talk their way through a daily note gets recommended a plugin that requires them to record a clip, stop, and wait for an API round trip. It works exactly as documented, and it is not what they wanted.

So the useful version of the question is not what is the best voice to text plugin for Obsidian. It is do you want to talk while you write, or process a recording afterwards, and then, separately, is Obsidian the only place you need this. Those two answers pick the tool between them.

What each recommended plugin actually does

These are the four plugins that come up most often for this query, including in Google's own AI summary of it. Every entry below is from the plugin's own listing or README rather than from a review of it, because the difference between the two turns out to matter.

Read the last column first. Three of the four do not do the thing most people are asking for. They are transcription tools, correctly built and clearly documented, that have been collectively filed under a word that means something else.

What the bring-your-own-key route costs

The three key-requiring plugins are all free as software, which is how they get described as free. The setup they actually ask for is worth spelling out, because it is usually presented as a single step and it is not:

• Create a developer account with OpenAI, Deepgram, Groq or Azure.
• Add a payment method to it, since these APIs are prepaid or metered rather than free.
• Generate an API credential and paste it into a note-taking application.
• Keep an eye on the balance, because the failure mode when it runs out is a note that silently does not get its text.

The per-minute rates are genuinely small. Speech to Text's own listing quotes Deepgram Nova-3 at $0.0043/min, which is a few cents an hour. The friction is not the money, it is the four steps and the ongoing responsibility for a billing relationship that exists so you can talk to your own notes.

No key, no billing, no plugin

Dictate into Obsidian without setting up an API account

BlabbyAI runs at the operating-system level, so the same shortcut types into an Obsidian note, your email, your browser and your chat. Custom modes turn a spoken thought into bullets or a task list as it arrives. Windows desktop app and Chrome extension. 60 credits a week free, no card.

Add BlabbyAI to Chrome

The Win+H question, and why every answer you have read is guessing

If you are on Windows, the obvious thought is to skip plugins entirely and use the voice typing built into the operating system. Search for whether that works in Obsidian and you will find confident answers in both directions. Having gone through the primary sources, here is what is actually established, because the honest answer is more useful than a confident one.

The famous quote is about macOS, and it is from 2021

Nearly every claim that dictation fails in Obsidian traces back to one forum thread. Here is its opening post, in full, because the framing matters:

Mac OS dictation feature. This is a macOS report, from January 2021, and it says nothing whatsoever about Windows. It gets quoted as though it were a general statement about Obsidian, and Windows users read it and conclude their own case is settled. It is not.

The replies under it also disagree with each other, which is more informative than a clean failure would have been. One user reports it working “around 10% of the time in Obsidian and 100% of the time everywhere else”. Another says they get dictation in Obsidian just fine. A third describes it working once or twice and then stopping. A separate bug report from November 2020 describes a long delay during which Obsidian appears unresponsive, and it was archived without a documented resolution.

The thread cited as Windows evidence is about something else

There is a 2024 Obsidian forum thread that gets linked as proof Win+H fails. Read it and it is about remapping the shortcut, not about dictation not working. The author says they use Windows voice dictation a lot and that it is activated with Win+H, and their complaint is that PowerToys and AutoHotkey cannot rebind that shortcut to another key inside Obsidian while the same rebinding works in Notepad and Word. That is a keyboard-hook problem in a desktop application window. It is not evidence about whether dictated text reaches the note.

The nearest thing to a direct first-hand Windows comment on the forum criticizes Win+H for being weak on accuracy and for supporting few languages besides English. That is a complaint about quality, which implies the thing runs, but implication is not confirmation and this page is not going to present it as one.

Microsoft does not settle it either

Microsoft's own documentation for voice typing requires only that you have your cursor in a text box, plus a microphone and an internet connection. They publish no application compatibility list and make no “works in every app” guarantee, so their documentation cannot be used to prove the Obsidian case in either direction.

The reason to lay this out rather than pick a side is that it points at the real decision. If a dictation route can be silently unreliable inside one specific application, the durable fix is a tool that does not depend on that application at all. BlabbyAI has its own Windows desktop app and puts text into whichever field has focus, so its behavior in Obsidian is the same as its behavior in Word, in a browser or in a terminal, rather than depending on how one application handles system dictation.

Why your phone already does this and your laptop does not

Worth one short section, because the contrast explains the whole desktop problem in a sentence. On a phone, the keyboard does the dictation, not the app. You tap the microphone on the keyboard, the system transcribes, and the keyboard inserts characters as though you had tapped them, so Obsidian is never in the audio path and has nothing to get wrong.

Desktop has no equivalent universal keyboard layer, which is precisely why the desktop version of this question is a mess and the phone version is not. That missing layer is the job a system-level dictation tool does on a computer. BlabbyAI puts the text into the focused field itself rather than asking the application to cooperate, which is the same mechanism your phone keyboard uses and the reason it does not care which window is in front. It is a Windows desktop app and a Chrome extension, so this is the desktop answer rather than a phone one.

The question that actually picks your tool

Every option so far has been scoped to Obsidian. Before settling on one, it is worth asking how much of your writing day Obsidian actually is.

For most people who keep a personal knowledge base, the notes are the part they think of as writing, and they are a minority of the words they type. The rest is email, chat, issue trackers, documents, browser forms, commit messages and search boxes. A plugin, by definition, covers none of that. Install Speech Kit and you can talk to your notes and then go back to typing everything else. If you are weighing that trade, our comparison of the best dictation apps covers the system-level options side by side, and whether dictation is actually faster than typing goes through what the studies measured.

Win+H is in that table honestly: it is free, it is already installed, and it is system-wide. Its weaknesses are the ones its own users name, which are accuracy on unusual vocabulary and thin language support, plus the unresolved question above about how it behaves in this particular application. For a note full of proper nouns, book titles, technical terms and the names of people you know, accuracy on unusual vocabulary is close to the entire job. There is more detail on that option in our piece on Windows voice typing, if you want to know what it handles before you test it.

Getting notes that are filed, not just captured

There is one more gap between transcription and a usable note, and it is the one that decides whether dictated notes actually get used. Speech, transcribed faithfully, is a wall of text. Notes are useful because they have shape: a heading, some bullets, a task, a link to another note. Cleaning that up by hand afterwards gives back the time the dictation saved.

BlabbyAI custom modes are free-form AI instructions applied to what you said, so the text can arrive already shaped. You pick the mode before you speak. Three examples, with the spoken input on the left:

**Pricing call**

- Interested in enterprise tier
- Pushed back on the seat minimum
- Marcus sending current contract this week
- [ ] Revised quote before the 15th

Spent most of today on the migration script. The tricky part: old records have no created timestamp, so I inferred it from the ID sequence. It works, but I am not confident about rows from before the 2019 import.

Chapter 4 argues attention is a finite resource, and that common interface designs are optimized to consume it.

Connects to: [[Notification batching]]

Because a mode is a free-form instruction rather than a fixed feature, it also covers spoken commands you define yourself. If you want saying “new paragraph” to break the line rather than type the words, that is a line in the mode. If you want technical vocabulary left strictly alone, that is another.

A setup that does not depend on one app

Given everything above, here is the order that makes sense for most people, and the reasoning behind each step.

• Start with the tool that is not tied to one app. Notes are rarely the only thing you write, so the shortcut that works in Obsidian should also work in your email, your browser and your chat. This is where BlabbyAI fits, and the free tier of 60 credits a week is enough to test it on your own writing before paying for anything.
• Set up two or three custom modes before judging it. Raw accuracy differences between modern models are smaller than the difference between a wall of text and a formatted note, and modes are what close that gap. One for daily notes, one for recaps, one that leaves technical vocabulary alone.
• If Obsidian genuinely is the only place you write, Speech Kit covers that case. Free, local, no key. It is the best of the plugins, and it is worth knowing it exists. For most people Obsidian is one window among several, which is the whole reason the first step is a system-level tool.
• Leave the free system dictation switched on. Win+H on Windows costs nothing to have available and is a reasonable backstop for a machine you have not set anything up on yet. Test it with a real paragraph rather than a test sentence, using the vocabulary you actually write, and you will know in a minute whether accuracy is the blocker for you.

Common questions

Sources

• Obsidian Forum, Dictation - speech to text, January 2021 to March 2023, for the macOS reports and the quoted opening post.
• Obsidian Forum, System dictation not working correctly, November 2020, for the delay-and-unresponsive report archived without a resolution.
• Obsidian Forum, Windows key remapping with PowerToys, December 2024, for the shortcut-rebinding issue that is widely misread as a dictation failure.
• Obsidian Forum, Voice input on Obsidian Mobile, March 2023, and the SwiftKey bug thread, for mobile behavior.
• Obsidian community plugin listings for Speech to Text, Speech Kit and Voice MD, for API key requirements, pricing and live-versus-batch behavior.
• The whisper-obsidian-plugin README, for its supported endpoints and the Alt+Q record-and-stop workflow.
• Microsoft Support, Use voice typing to talk instead of type, for the stated requirements of Win+H.
• Obsidian developer documentation, Mobile development, for the confirmation that the desktop application exposes Electron APIs.