Speech To Markdown

• 1. overview (684 chars)

  Speech To Markdown is a free and open source macOS/iOS application created by independent developer Igor Steblii. The core concept is "Speak, get clean Markdown, 100% local processing." It uses local Whisper speech recognition + local LLM real-time structuring to directly convert spoken content into well-formatted Markdown documents. In a market environment where competing products generally charge an annual fee of $144 and rely on the cloud, Speech To Markdown provides a differentiated path for privacy-sensitive technology users with the triple advantages of being completely free, 100% offline, and open source code. The product is currently in the early stage (v0.2.0) and the community popularity has not yet formed, but the underlying logic and user experience direction have been recognized.

• 2. Product background (1220 chars)

  Speech To Markdown (stmd for short) was developed by independent developer Igor Steblii (GitHub: xajik). Igor is a full-stack developer. He revealed on LinkedIn that his immediate motivation for developing this application was: "The results of talking to Gemini became worse and worse, and finally gave up and went back to typing. Voice input is indeed faster than typing, but the real challenge is not speed, but structure - it is difficult to organize your thoughts while impromptu discussing complex processes, products or technical requirements."   The development process of the application began in 2025. It was initially iterated on GitHub under the name VoiceToMarkdown and has evolved through four versions: 0.0.4→0.0.8→0.1.0→0.1.1. On July 10, 2026, the project was renamed SpeechToMarkdown and version v0.2.0 was released. At the same time, the iOS version of SpeechToMarkdown is available on the App Store, supporting iPhone 15 Pro and above. As of now, the GitHub repository has 41 commits, 2 contributors (including Claude AI assisted contribution), and the App Store application is only 3.1 MB.   The positioning of Speech To Markdown is very clear: it is not a mass consumer product, but an efficiency tool for technical users, developers, and heavy Markdown users. It does not pursue cross-platform coverage or cloud synchronization functions, but achieves the ultimate in depth, accuracy and privacy within the Apple ecosystem.

• 3. Product functions and usage experience (2546 chars)

  Speech To Markdown provides two versions - macOS desktop version and iOS mobile version. The two sets of technology stacks are different but the experience is the same.   Let’s talk about the macOS desktop version first. It operates in the menu bar and can be invoked at any time through the shortcut key ⌘⌥] - the "global dictation" mode is activated in any application and the transcribed Markdown text is directly entered at the current cursor position after speaking. This "global injection" method breaks through the input barriers of all apps, and you can use it in terminals, browsers, Slack, and VS Code at will. In addition to global dictation, there is also "agent mode" - a real-time Markdown editor window. When you speak, your voice is first transcribed into text by whisper.cpp (buffered every 4 seconds), and then sent to the local LLM for structured processing. The results are streamed into the editor, and what you see is what you get.   The iOS mobile version has taken a completely different technical route. It does not rely on any external LLM server, but calls the built-in SpeechAnalyzer of Apple devices for speech recognition and Apple Foundation Models for formatting. This means it has zero configuration, zero dependencies, and even supports airplane mode. Since the launch of the App Store, users only need to download and use it. No registration, no settings, and no network connection are required.   The most noteworthy thing is its three working modes. Format mode is a free dictation mode. What you say will be sent to LLM to reconstruct the entire document in real time - all new content is sent to the model along with existing content to ensure global consistency. Edit mode treats voice as instructions rather than content, such as "Change the subtitle to Weekly Notes" and "Convert this list to a table" - select the text and speak the instruction, and the model only changes the selected part. Append mode is an acceleration mode designed for long documents. Only the last three sentences plus new content are sent to LLM, and the delay does not increase with the length of the document.   The output format supports Markdown, plain text and HTML, which can be switched at any time during the session, and the settings are retained across startups. All sessions are automatically saved and the original transcript is always available with one click.In terms of competitive product comparison, Typeless ($144/year, cloud processing) and Wispr Flow ($144/year, cloud processing) provide a more mature cross-platform experience, but they are both paid subscriptions and rely on the cloud. Trace (£9.99 one-off buyout, macOS only) is more professional at meeting transcription and speaker separation, but lacks real-time structuring capabilities. v2md (starting at $14.49/year, offline on iOS) is the closest competitor, but its charging model and positioning are controversial. Speech To Markdown has no rivals in the four dimensions of "completely free + open source + 100% offline + native Markdown support".

• 4. Pricing and business model (947 chars)

  Speech To Markdown is currently completely free, with no in-app purchases, no subscriptions, and no ads. The iOS version is available on the App Store, and the macOS version is distributed via GitHub. The code is open source on GitHub and anyone can freely download, compile and modify it.   This model is extremely rare among similar products. Typeless has an annual fee of $144, Wispr Flow has an annual fee of $144, Brain Dump has an annual fee of $44.99, and v2md has an annual fee of $14.49–$35.99. The only one that's close to free is Trace (£9.99 one-off buyout), but the functionality is completely different. Igor did not disclose future monetization plans, but it can be speculated that he may achieve long-term sustainability through sponsorship (such as GitHub Sponsors) or additional services (such as cloud synchronization, team version).   Currently stmd does not have enough reviews on the App Store to display a rating, and the number of stars on GitHub is also in its infancy. But this tool shows a software path that is different from SaaS subscription: small but beautiful, free and open source, and technology-driven.

• 5. User reviews (511 chars)

  Since Speech To Markdown has only been on the App Store for a short time, it has not yet accumulated enough user reviews.   But there are already some voices in the developer community. Igor published an article titled "From Brain Dump to Markdown: Structure Ideas as You Speak" on DEV.to, dictating the entire article using stmd as a demonstration. There are also seed users on LinkedIn who said, "The experience is unexpectedly good - I just said something casually, and it not only converted it into text, but also directly organized it into Markdown for me and automatically split it into clear 1, 2, and 3 points."

• 6. Industry evaluation (988 chars)

  Currently, this app has received very little exposure in industry media, with no reports from mainstream technology media such as TechCrunch and The Verge. But in the developer tools community, tool navigation sites such as thistools.app and gunbark.dev have picked it up and reviewed it favorably. The review of thistools.app states: "The underlying logic of this architecture is very clear - the audio is transcribed and buffered by whisper.cpp in sections of about 4 seconds, and when it accumulates about 30 words, it is sent to LLM, and the results are streamed to the editor. This design achieves a good balance between latency and global consistency."   In terms of competitive product landscape, the market is expanding rapidly. Typeless just received a new round of attention in July 2026. Wispr Flow has raised $81M and is valued at $700 million. The speech → structured text track is changing from "a toy for a few people" to "an infrastructure that everyone needs." Speech To Markdown is free and open source. Although it is difficult to form commercial competition in the short term, it has a solid niche market in the technical circle and privacy-sensitive user groups.

• 7. Controversies and Risks (870 chars)

  As a free open source tool, the biggest risk Speech To Markdown faces is not commercial competition, but developer energy and project sustainability. Independent developer projects often follow the trajectory of "enthusiastic start → rapid iteration → slow stagnation". If Igor cannot continue to invest in maintenance, stmd may gradually become inactive like many excellent open source projects.   Another risk is hardware limitations. The iOS version requires iOS 26+ and an Apple Intelligence device (iPhone 15 Pro and above), which excludes a large number of existing users. The macOS version requires users to install whisper.cpp and the local LLM server by themselves, which has a higher threshold for non-technical users. In the Chinese ecosystem, this threshold is particularly prominent - the application interface only supports English, and there are almost no Chinese reviews or tutorials on China's mainstream platforms (Zhihu, Xiaohongshu, Bilibili, CSDN, etc.), which is a significant user growth bottleneck.

• 8. Applicable groups and usage suggestions (637 chars)

  Speech To Markdown is most suitable for three types of people: first, technical users and developers who are willing to spend time configuring local LLM in exchange for unlimited free use and the highest privacy guarantee; second, heavy users of Markdown note libraries such as Obsidian, who need an efficient dictation input method; third, users who are extremely sensitive to data privacy and hope that voice data never leaves the device.   People who are not suitable include: ordinary consumers looking for an out-of-the-box experience (Typeless or Wispr Flow is recommended), users who need cross-platform multi-device synchronization, and users who are looking for meeting transcription and speaker separation functions (Trace is recommended).

• 9. Summarize (435 chars)

  Speech To Markdown is a noteworthy new player in the speech-to-Markdown field. It chose an anti-mainstream path - no money, no Internet, no out-of-the-box use - but precisely because of this it filled a real gap. Independent developer Igor Steblii used a set of exquisite technical solutions (local Whisper + local LLM + three working modes) to prove that speech to structured text can be extremely private and completely free. For target users, it may not be a mature product, but it is definitely in the right direction.

No in-depth report yet

• 

  Robin749

  —

  Found a bottleneck — voice pickup is not great in noisy environments, since it uses the Mac's built-in microphone. Connecting an external mic helps a lot.

---

• 

  冯明

  —

  Chinese voice recognition accuracy is better than expected, using the Whisper model which handles Chinese well. However, the Markdown formatting style leans towards English conventions, and Chinese typography occasionally needs manual adjustment.

---

• 

  狗狗112

  —

  Compared to Wispr Flow, it lacks cross-platform sync and AI cleanup features, but it wins on being completely free with data staying on-device. Depends on personal priorities.

---

• 

  ElizabethJenkinsX455

  —

  Wish it supported Android, but looking at the architecture it depends on Apple frameworks, so probably not happening anytime soon.

---

• 

  陈丹丹

  —

  The streaming output of the real-time editor is cool. Watching the text appear line by line as you speak feels like you're writing code yourself (lol).

---

• 

  SHernandezQ

  —

  Product Hunt: 89 votes, ranked 15th. Pretty impressive for a newly released free open-source project. Hope to see continuous iteration.

---

• 

  黄云鹏

  —

  Tried using the iOS version on a plane. Flight mode works perfectly — data doesn't go to the cloud, which is a delight for frequent travelers.

---

• 

  TCastilloJr

  —

  Developer Igor wrote an entire article using this tool on DEV.to as a demonstration. Quite an interesting eat-your-own-dogfood case.

---

• 

  TheXavierScott

  —

  Would be perfect if speaker diarization could be added in the future. Right now only one person can use it, multi-person meeting scenarios aren't covered yet.

---

• 

  EDort

  —

  Hey guys, for the macOS version I chose Qwen 3.5 27B 4bit when configuring the local LLM, running with omlx, and the speed is decent. Has anyone tried Gemma 3? How is it?

---

• 

  realEmaRikstad_x

  —

  Used v2md for a few months. Seeing this one come out completely free is surprising. The features might not be as complete as v2md yet, but as open source the potential is huge.

---

• 

  DebraFosterSr

  —

  The global dictation hotkey setup is decent, doesn't conflict with Alfred or Raycast. But the first launch requires microphone and accessibility permissions, which might be off-putting for beginners.

---

• 

  Natalie_Ward_77

  —

  The natural language understanding of commands in edit mode is not stable enough, but considering it's a free and open-source early-stage product, it's already impressive.

---

• 

  Jessica_Kelly_20227

  —

  Tried it during a meeting. Just dictated the meeting notes verbally and it converted them into structured Markdown. Way more efficient than taking notes manually.

---

• 

  jcmbo8rhv777

  —

  After comparing several speech-to-Markdown tools, this one strikes the best balance between privacy and functionality. No registration, no internet required, no payment — what more could you ask for?

---

• 

  LaurenCruzSr

  —

  Three output formats are very practical: Markdown for documentation, plain text for quick notes, and HTML for direct preview. Switching between them is smooth.

---

• 

  TerryRiveraJr

  —

  Tried Gemma 3 and Qwen 3.5. Personally, I feel Qwen's formatting quality is better, while Gemma wins on speed.

---

• 

  crazyswan128

  —

  Looked at the source code on GitHub. Igor's code quality is good. But the project is still early, only 41 commits — I'm a bit concerned about whether maintenance can keep up.

---

• 

  itjpsxl6

  —

  Tried Edit mode. Saying 'change the subtitle to Weekly Notes' actually worked — a keyboard-free editing experience that feels magical. But precision still has room for improvement; complex instructions are sometimes misunderstood.

---

• 

  Donald.GrayZ383

  —

  Append mode is incredibly useful. When writing long documents, the latency doesn't increase as the content grows — it only appends new content. This is a very smart design.

---

• 

  JohanMortensen

  —

  Compared to Typeless, this one wins on being completely free and processing data locally. But Typeless's out-of-the-box experience is indeed better. Different strokes for different folks.

---

• 

  孔飞梅

  —

  Downloaded the iOS version, amazed by the 3.1 MB size. But it requires iOS 26 and iPhone 15 Pro or above — my 14 Pro is precisely excluded. Crying.

---

• 

  DylanHicks_99

  —

  Downloaded and tried the macOS version. The setup process does have a learning curve for regular users — you need to install whisper.cpp and a local LLM first. But once configured, the experience is amazing: speak and get structured Markdown output directly, and the global dictation hotkey works in any app.

---

• 

  ticklishswan803

  —

  I've been looking for an Obsidian voice input solution, and this one outputs .md files directly, perfect match. Free and open source — amazing.

---

• 

  CAhil

  —

  Saw Speech To Markdown on Product Hunt. Free and open-source, gotta support it. Looked at the GitHub repo, the architecture is clear: local Whisper for transcription, local LLM for structured output, data never leaves the machine. This level of privacy blows those cloud solutions out of the water.

No reviews yet