What you say
Um, about next week’s meeting: Tuesday—no, Wednesday at 11 a.m., please. Also, it would help if you could share the materials first.
AI dictation for Mac Beta
All on your Mac.
Turn spoken corrections and fillers into clear emails and notes, right at your cursor. Recognition and text cleanup can both run on your Mac.
No card required. No automatic charge.
v1.4.21 · · See release history
Seven days from your first trial activation. Internet is required during the trial; a purchased licence is required afterwards.
Apple Silicon / macOS 26+
We are continuing to improve this beta. For issues or questions, contact us.
Press Right Option again to insert text.
Dictation result
Um, about next week's meeting, Tuesday—no, please move it to Wednesday at 11 a.m. Also, please share the materials in advance.
Please move next week's meeting to Wednesday at 11 a.m. Please also share the materials in advance.
Orqnote
What you say
Um, about next week’s meeting: Tuesday—no, Wednesday at 11 a.m., please. Also, it would help if you could share the materials first.
What gets inserted
Please move next week’s meeting to Wednesday at 11 a.m.
Please also share the materials in advance.
Speak your tasks in order to create a readable list.
What you say
Three tasks: first, check invoices; second, enter expenses; third, plan next month.
What gets inserted
Three tasks:
• Check invoices
• Enter expenses
• Plan next month
With screen-aware formatting enabled, Orqnote uses the text visible in your destination app as a guide: polite for emails or casual for chats. Results depend on what it can read. When disabled, it generally preserves your spoken tone.
What you say
I need a reply by tomorrow morning. If that is not possible, tell me today.
What gets inserted
Please reply by tomorrow morning. If that is not possible, please let me know today.
Register pronunciations and spellings in your dictionary to reduce corrections to names and product terms.
What you say
Send the Orqnote quote to Ms. Watanabe.
Example with registered spellings
Send the Orqnote quote to Ms. Watanabe.
Illustrative sample. Results vary with settings and speech.
Clean up corrections and fillers, with punctuation and line breaks.
Use a personal dictionary, correction learning, and optional screen context for names and terminology.
On-device mode processes your words without sending them to external AI, with no metered AI fees.
Email replies / Chat / Notes / Blog drafts
Place your cursor in a text field.
Press Right Option and speak.
Press again to insert your text.
You can change the shortcut. If no input field is found, the text is copied.
Why local AI?
On-device mode keeps audio and text away from external AI. After purchase, registration and model setup, it also works offline. The trial requires internet access to verify its expiry.
For general cloud AI
Audio or text is sent to the selected service. Storage and data handling depend on its terms.
Orqnote on-device mode
This is a configuration that performs voice recognition and text formatting within the Mac. No need to send to external AI. The choice is yours.
Transcription is done on your Mac, and only the transcribed text is sent to the API. The image of the front window is also sent only when "Formatting by looking at the screen" is turned on. What is sent is different from the general cloud AI diagram above.
When network connections are used.
Even in on-device mode, communication occurs in the following situations.
Audio, transcription, and formatting can be completed within the Mac
Choose when to use external AI.
Internal processing is designed to reduce the amount of information sent to external services. However, security measures for the Mac itself (OS updates, screen lock, access privilege management) are required separately. Please only give permission for microphone and screen recording when necessary. Always exclude password management apps from screen capture. We do not perform app-specific access analysis.
Pricing & purchase
A one-time ¥1,980 purchase, including tax. On-device mode has no metered AI fees.
Sales price
¥1,980
Tax included ・ One-time purchase
Secure payment with StripeEnter your card details directly with Stripe. This site does not store them.
Orqnote is an independently developed Mac app. It is not yet signed with an Apple Developer ID or notarized, so when you first open the downloaded Orqnote.app (by double-clicking), macOS shows a warning that the developer cannot be verified.
We are aware of this warning. The Orqnote we distribute does not contain malware. “Cannot be verified” means Apple has not confirmed the developer. It does not mean that a virus was found.
Removing this warning requires an Apple Developer Program membership (paid, annual) plus signing and notarization. As an independent developer, we cannot cover the cost up front; we plan to do this once sales reach a certain level. The timing is not decided yet. Until then, we ask for a one-time setup on first launch.
You do not need to disable security for the whole Mac. See the steps with screenshots / Apple Support: Open a Mac app safely
On sale as a beta. Change-of-mind refunds are generally unavailable. See the terms for statutory rights and remedies for non-conforming products. Terms
Ask before buyingQuestions before buying? Contact.
Typeless Pro costs about ¥4,800 per month on monthly billing, or ¥57,600 over a year. Orqnote costs ¥1,980 once, including tax, for continued use in on-device mode.
| Service / billing | Monthly cost | 1-year total | 3-year total |
|---|---|---|---|
| Typeless Pro (monthly payment) | About ¥4,800/month $30/month | About ¥57,600 $360 | About ¥172,800 $1,080 |
| Typeless Pro (annual payment) | About ¥1,920/month equivalent $12/month, billed annually | About ¥23,040 $144 | About ¥69,120 $432 |
| Aqua Voice Pro (monthly payment) | About ¥1,920/month $12/month | About ¥23,040 $144 | About ¥69,120 $432 |
| Aqua Voice Pro (annual payment) | About ¥1,600/month equivalent $10/month, billed annually | About ¥19,200 $120 | About ¥57,600 $360 |
| Wispr Flow Pro (monthly billing) | About ¥2,400/month $15/month | About ¥28,800 $180 | About ¥86,400 $540 |
| Wispr Flow Pro (annual billing) | About ¥1,920/month equivalent $12/month, billed annually | About ¥23,040 $144 | About ¥69,120 $432 |
| Orqnote(One-time purchase) | No monthly fee ¥1,980 once (tax included) | 1,980 yen(Tax included) | 1,980 yen(Tax included) |
The monthly equivalent of an annual plan is not a monthly payment. The ¥1,920 monthly equivalent for Typeless Pro and Wispr Flow Pro requires about ¥23,040 paid annually. The ¥1,600 monthly equivalent for Aqua Voice Pro requires about ¥19,200 paid annually.
Yen figures use an indicative rate of US$1 = ¥160. Competitor taxes, exchange-rate changes and payment fees are excluded. Comparisons use standard individual Pro prices on official websites, excluding free plans, discounts and app-store pricing. Features and terms differ. Totals assume unchanged prices. Orqnote assumes continued use of the purchased version, with no additional AI fees in on-device mode. External API usage is charged separately.
Pricing sources:Typeless official pricing / Aqua Voice official pricing / Wispr Flow official pricing
Operating environment and initial settings
Requires Apple Silicon and macOS 26 or later. Developed on a 32GB Mac. If memory is insufficient, consider API-based formatting or Apple Intelligence.
The transcription model is approximately 1.5GB, and the 9B model in the terminal is approximately 5.6GB. Please check the free space and communication amount as it will be obtained when using it for the first time. Once obtained, the configuration allows input in on-device mode without a network.
Because the app is not signed or notarized by Apple, macOS shows a warning on first launch. The Orqnote we distribute does not contain malware. Follow “If the first-launch warning appears” below and allow it once. On managed Macs, an administrator may need to approve it.
Microphone, input monitoring, accessibility. If you use formatting by looking at the screen, record the screen. The permission screen can be opened from "Access Privileges" in the menu.
Open the ZIP you received after purchase.
Move Orqnote.app to the "Applications" folder.
Register this Mac using the key and email confirmation code from your purchase email.
Double-click Orqnote.app and macOS shows a warning. Follow the first-launch steps to allow it, then allow the required permissions.
Obtained during the first transcription. If you use on-device formatting, obtain it additionally.
Place the cursor in the input field and press the right Option to speak.
There is no need to install Python separately. API key registration is required only when using API mode.
Only on the first launch, macOS asks you to allow the app. The warning appears when you first open the downloaded Orqnote. Follow the screens in order; the setting location is included.
Unzip the download, put Orqnote.app in your Applications folder, and double-click it. Because the app is not signed or notarized yet, macOS shows a warning on this first launch. The warning is expected.
Do not choose “Move to Trash”. Leave Orqnote.app in your Applications folder.
Apple menu → System Settings → Privacy & Security in the sidebar → scroll down to the Security section.
In the same Security section, macOS shows that “Orqnote.app” was blocked because the developer cannot be verified. Choose “Open Anyway”, confirm with your Mac password or Touch ID, and choose “Open” once more.
If you cannot find “Open Anyway”, close the warning, try to open Orqnote again, then check the setting within one hour.
This is only needed the first time. You do not need to lower security for the whole Mac. Reinstalling the app, or some updates, may require the same confirmation again.
If the warning says the app is damaged, or reports malware, do not open it. Contact the address in your purchase email. The sender is listed there, and you can also verify the ZIP’s SHA-256.
Things to check before purchasing.
Orqnote is currently sold as a beta release. We are still adding and improving features while people use it for everyday input. Your purchased version remains usable. If you find a bug or something unclear, use the contact form and choose “Bug report.” Do not include API keys, license keys, or other secrets.
If you use on-device mode, there is no need for a contract such as ChatGPT. When using API mode, a contract such as OpenAI API, API key, and usage fee are required separately. ChatGPT's paid plan and API usage fees are separate.
The audio itself is not sent to the API. Transcription is done on your Mac. On-device mode also does not send the transcribed text to external AI. In API mode, the transcription text is sent to the configured API. If "Formatting by looking at the screen" is on, the image of the front window will also be sent. We do not perform app-specific access analysis.
Requires an Apple Silicon Mac and macOS 26 or later. Developed on a 32GB Mac. If memory is insufficient, consider API-based formatting or Apple Intelligence. The app is about 1GB (download ZIP: about 352MB), the transcription model about 1.5 GB and the on-device formatting model about 5.6 GB. Tested on an Apple M4 with 32 GB of memory.
In on-device mode, the app does not set an upper limit on the number of inputs. Processing time depends on the length of the recording and the performance of your Mac. By default, up to 100 history items are retained. API mode restrictions are subject to the terms of the selected service.
Refund terms are listed on the purchase page and inTerms of use. Statutory rights and remedies for product defects are unaffected. For details, seeTerms of usePlease check.
Try free for 7 days. Download the app and choose Start free trial during setup. No email or card is required, and there are no automatic charges. Internet access is required during the trial. Each Mac gets one trial; reinstalling does not extend it. Register a purchased licence to continue afterwards. External API services charge their own usage fees during the trial too.
AI text cleanup
Don't just ``transcribe it as is'', create sentences that are easy for readers to understand. Adjust rephrasing, fillers, punctuation, line breaks, and bullet points to the extent that they do not change the intent of the utterance.
Before cleanup
The next meeting will be on Monday, oh no, Tuesday from 3pm.
After cleanup
Please schedule our next meeting for Tuesday at 3 p.m..
If you are unsure of the word, go to the last word you chose. Consolidate phrases like "Monday, no, Tuesday" into a natural sentence.
Before cleanup
Um, Um,This is new material, butThat,Share by Friday.
After cleanup
We will share new materials by Friday.
Remove fillers such as "um," "um," and "that" from your sentences. Prioritize readability without retaining too much of the quirks of spoken language.
Before cleanup
There are three topics on today's agenda. First, check your budget. Next is the schedule. The last step is to assign the person in charge.
After cleanup
There are three topics on today's agenda.
First, check your budget. Next is the schedule. The last step is to assign the person in charge.
In addition to periods, insert paragraphs and line breaks to match the meaning. By default, each sentence is wrapped at a readable position.
Before cleanup
There are three things to do. The first step is to check your invoice, the second step is to enter your expenses, and the third step is to summarize next month's schedule.
After cleanup
There are three things to do.
・Confirmation of invoice
・Enter expenses
・Summarize next month's schedule
Pick up cues to enumerate, such as "the first one" and "the second one," and write them down in bullet points. Make it readable without having to correct it by hand.
Before cleanup
I will send Mr. Watanabe a quote for Orc Note.
After cleanup
Mr. WatanabeTo,OrqnoteI will send you a quote.
Correct variant spellings of people's names and product names using dictionary and on-screen terms. If you register the reading, it will be reflected from the next input.
Before cleanup
I would like a reply on this matter by tomorrow morning. If that is not possible, please contact us today.
After cleanup
This matter isuntil tomorrow morningI need a reply. If it is difficult,TodayPlease contact us.
Don't over-summarize or add information that isn't in the utterance. Adjust honorific language and phrasing as long as it does not change the meaning.
Shaping involves guessing by AI. Please check the contents of important documents before sending them. Results vary depending on model, setting, and speaking style.
Screen-aware text cleanup
When you start recording, read the terms and notations from the front window. It assists in recognition and formatting using clues from product names, people's names, and technical terms appearing on the screen.
meeting memo
・Mr. Sasaki(Minato Shoji)
・Orqnote implementation cost
・Qwen3.5 Verification results of
・MLX operation in
Use the terms on the screen as hints for reading.
Look at the screen and format: On
Before cleanup
I will send the Orknote installation cost and the verification results of Kyuen 3.5 to Mr. Sasaki.
After cleanup
Mr. SasakiTo,Orqnoteintroduction cost andQwen3.5I will send you the verification results.
Terms that match the notation on the screen are used as a reference for recognition and formatting. The terminology on the screen is not always adopted.
If you switch to API mode with "Formatting by looking at the screen" on, the image of the front window will also be sent to the set API. Before changing the mode, check "AI used for formatting" and destination in the menu.
Dictionary and correction learning
People's names, product names, company terminology. You can also memorize and correct words that are difficult for general speech recognition using your own dictionary and corrections.
Register the words to be read and the notation you want to input. You can specify the reading, such as "Saito" → "Saito".
saitōSaitō (preferred spelling)
If you have just entered the text, you can register it by selecting "Dictionary" → "Register from last input..." and leaving only the necessary words from the sentence.
orukunootoOrqnote
When you correct the text you entered and start the next voice input, you will see the corrected words. The words you have memorized will be displayed in "Memorized Phrases".
person name on screenTo the correct notation
Unnecessary words can be deleted individually or all at once. Dictionaries, history, and memorized words are saved on your Mac.
Keep history100 items by default
For corrections read from images, check the same combination twice before using it as a reference for corrections. If it does not memorize automatically, please use "Register from last input...". If it has been more than 30 minutes since the modification, it may not be detected.
You can choose the device and API
The AI used for text formatting can be switched from the menu. The transmission range and cost vary, so you can choose according to the content of your work.
On-device (MLX)
This is the most ``keep it out'' configuration. For daily input, start with this.
API
When you want to choose the speed, model, and price yourself. If you look at the screen and formatting is on, the image will also be sent.
Not used (transcription only)
First, when you want to check the raw transcription.
In API mode, the transcription text is sent to the configured API. If "Format by looking at the screen" is on, the image of the front window will also be sent. You can check the destination under "AI used for formatting" in the menu.
| What to send | On-device (MLX) | API |
|---|---|---|
| audio | Do not send (transcribe within Mac) | Do not send (transcribe within Mac) |
| transcribed text | do not send | Send to set API |
| screen image | Do not send (processed within Mac) | Send when looking at the screen and formatting is on |
| API key | Unnecessary | Save to macOS Keychain |
| AI usage fee | No additions | Price of selected service |
API keys are stored in the macOS Keychain and not in configuration files. The audio itself is not sent to the API, and transcription is always performed on the Mac. If you select "Do not use", the transcription will be input without performing AI formatting. API data handling is subject to the terms of the selected service.
Comparison with Typeless/Aqua Voice (confirmed on October 26, 2026)
Voice input options have increased. Both Typeless and Aqua Voice have strong cloud models and sophisticated experiences. What's different about Orqnote is that you can perform speech recognition and text formatting on your Mac. You can choose the processing location and payment method.
In internal on-device mode, neither audio nor text is sent to external AI. Use the API only when necessary.
Transcription and formatting in on-device mode is done within the Mac. API fees do not increase depending on the number of conversations.
You can run the 9B model on your device and choose the model and API yourself.
| Feature | OrqnoteFor Mac | Typeless ProCloud-based | Aqua Voice ProCloud-based |
|---|---|---|---|
| Speech processing | Inside Mac (Whisper) | Cloud (policy: audio is not retained after processing) | Cloud (no saving in Privacy Mode) |
| AI text formatting | Yes: choose on-device or API | Yes (cloud) | Yes (cloud) |
| On-device LLM cleanup | Yes (Qwen3.5 9B) | Not stated in official sources | Not stated in official sources |
| Offline dictation | Available after downloading models (on-device mode) | Not stated in official sources | Not stated in official sources |
| Dictionary | Yes (dictionary + corrective learning) | Yes | Yes |
| screen context | Yes (terminology for front window) | Tone adjustment for each app (reading of screen content is not mentioned in the official information) | Yes (Deep Context) |
| Personal pricing | ¥1,980(Tax included/One-time purchase) | Monthly billing: about ¥4,800/month ($30) Annual billing: about ¥1,920/month equivalent ($12) About ¥23,040 paid annually | Monthly billing: about ¥1,920/month ($12) Annual billing: about ¥1,600/month equivalent ($10) About ¥19,200 paid annually |
| Free plan | 7-day free trial | Yes (limited number of words) | Yes (1,000 words) |
| One-time purchase | One-time purchase | Subscription | Subscription |
| Compatible OS | macOS 26 or later (Apple Silicon) | macOS・Windows・Linux・iOS・Android | macOS・Windows・iOS・Android |
| Data handling | On-device mode is not sent to external AI. Select destination for API mode | Cloud processing. It is clearly stated that the audio will not be retained and will not be used for learning. | Cloud processing. No server storage in Privacy Mode |
Features checked on official websites as of October 26, 2026. Prices checked October 27, 2026, using US$1 = ¥160. Competitor taxes, exchange-rate changes and payment fees are excluded. Typeless’s free plan has a word limit (no amount stated on its official page); Aqua Voice includes 1,000 words. Plans, prices and features may change. “Not stated” means the reviewed official information does not specify it, not that it is unsupported. Orqnote reflects the evaluation build’s implementation, not a side-by-side device test. Sources:Typeless price、Typeless product information、Aqua Voice Price、Aqua Voice product information。
Typeless and Aqua Voice have big cloud models, sophisticated UX, free plans, and cross-platform. Orqnote is an option for those who prioritize processing location and cost structure. The standard macOS voice input is free and easy, but the functions of AI text formatting, on-device LLM, and API selection are different.
Distribution and updates
You can register with the license key from the purchase email and check and apply the new version from within the app. It can be used on a Mac owned by the purchaser, and there is no deadline for official updates.
distribution
Provides distribution ZIP and SHA-256. When starting up for the first time, you need to confirm the macOS.
Registration
Register using the license key from the purchase email and the confirmation code sent to the email address at the time of purchase. You can also register other Macs that you own using the same key, and there is no limit to the number of them.
update
You can check for new versions, verify and apply signatures by going to Updates in the app. Application will be postponed while recording, transcription, and formatting are being processed. There is no set period for providing updates.
Settings and data
Even if you update the app, settings, dictionaries, and history will remain on your Mac. Store your API key and license information in Keychain.
I don't write about what I can't do as if I did.
It is necessary. Enter the license key from the purchase email and the email address at the time of purchase on the "License" screen of the app, and confirm your identity using the confirmation code you receive. You can't register a new Mac with the key alone. An unlimited number of Macs owned by the purchaser can be registered. The API key for API mode is separate from the license key.
This is an AI voice input app for Mac. When you speak into the input field where the cursor is located, it will perform transcription and AI text formatting and input. Speech recognition is performed using Whisper on the Mac, and text formatting can be selected from the device model or API.
Typeless and Aqua Voice have big cloud models, sophisticated UX, free plans, and cross-platform. Orqnote is different in that you can perform voice recognition and text formatting on your Mac, there is no pay-per-use charge for AI in on-device mode, and you can choose the model and API yourself. If the accuracy of the latest models in the cloud is your top priority, competition may be the way to go. For more detailsComparison tablePlease take a look.
Speech recognition always takes place within your Mac. Text formatting is done on the Mac when you select the internal model (Qwen3.5 9B). Communication occurs during initial model acquisition and API mode.
If you have acquired the model, on-device mode is a configuration that can be used without a network connection. The license retains the last confirmed usage rights for a certain period of time, so that it does not stop immediately even if there is a temporary communication failure. Connection is required for initial model acquisition, API mode, and update confirmation.
It is not something that can be directly compared. Orqnote has a 9B model installed in the terminal and performs verification including rewording and filler removal. As for recognition accuracy, the character error rate for 24 Japanese examples was 1.19% (synthesized speech). In another 14 cases, the median processing time for silence detection → transcription → formatting was 2.47 seconds. Both results are based on the development environment in October 2026, and will not be the same for all utterances or environments. If it does not suit your purpose, you can also choose API mode.
In testing in October 2026, the median time for silence detection → transcription → formatting was 2.47 seconds (14 cases, MacBook Air M4, 32GB, synthesized voice). Transcription takes 1.00 seconds, formatting takes 1.01 seconds. It does not include the process of stopping recording, capturing the screen, or pasting text. Actual time will vary depending on recording length, Mac performance, and settings.
Can be set. Register from "Register API key..." in the menu. The key is stored in the macOS Keychain and not in the configuration file. You can change the connection destination and model name in advanced settings.
It supports OpenAI compatible Chat Completions API. Please specify the base URL, model name, and API key. Depending on the specifications and model of the connected device, not all functions may be available.
Corrections can be made by registering in the dictionary, "registering from the last input...", corrective learning from rework, and "formatting by looking at the screen." You can personalize words that are difficult for general speech recognition, such as people's names, product names, and technical terms.
This function reads one sheet of the front window when recording starts, recognizes the terms and notations appearing on the screen, and uses them as a reference for formatting. Password management apps are always excluded, and you can also set excluded apps. You can turn it on or off from the menu. In API mode, images are also sent to the setting destination.
Dictionary entries and memorized phrases can be deleted individually from the menu. Memorized words can be deleted all at once. History is stored within the app and retention can be disabled in settings. Dictionaries, history, and memorized words are saved on your Mac.
Check, download and install new versions from Updates in the app. Update information and files are verified with signatures, and installation waits until processing finishes. Without Developer ID signing or notarization, macOS may require confirmation or renewed permissions after first launch or an update. This is due to our signing method, not a setup mistake by the purchaser. We plan to join the Apple Developer Program and obtain signing and notarization, but have no date yet. Purchasers do not need to register as developers. There is no set end date for updates.
Yes. The app is a one-time ¥1,980 purchase, including tax. On-device AI has no metered fees. Optional external APIs are billed separately.
Because Orqnote is not signed or notarized by Apple, double-clicking the downloaded app for the first time may show that the developer cannot be verified. The app we distribute does not contain malware. Choose Cancel, then allow it with “Open Anyway” in System Settings → Privacy & Security → Security.
You can receive it.Launch notificationPlease register your email address here. We will inform you only when the price and sales method are decided. Registration will be used for notification purposes only, and for consultation before purchase, please use the inquiry form.
Orqnote is for Mac only. Applies to Macs with Apple Silicon and macOS 26 or later; there are no Windows, iPhone, or Android versions. Not compatible with Intel Macs. Please inquire before purchasing if it will work in a virtual environment or on a managed Mac.
The recorded content will be saved on your Mac until the input results remain in the history. If you are unable to continue transcribing, you can check by selecting "Open saved recording..." from the menu. The recording will be deleted once normal entry is complete.
Dictate → transcribe → clean up
Speak naturally, including pauses and corrections. Whisper transcribes your speech on your Mac. A local LLM removes fillers, resolves corrections, and adds punctuation before inserting the text at your cursor.
00:06
Recording
STEP 1 / 4
"Well, the meeting is next week, but please reschedule it to Tuesday or Wednesday at 11 o'clock.Also, it would be helpful if you could share the materials first."
listening to audio
Um, about next week's meeting, Tuesday—no, WednesdayPlease change it to 11 o'clock.Also, Materials tooin advanceIt would be helpful if you could share it.
Please change next week's meeting to Wednesday at 11 o'clock.
It would also be helpful if you could share materials in advance.
Press Right Option and speak naturally.
Voice recognition is performed within the Mac. Text formatting is performed by the selected AI (on-device model or API).
STEP 1Speak
"Well, the meeting is next week, but please reschedule it to Tuesday or Wednesday at 11 o'clock.Also, it would be helpful if you could share the materials first."
Press right Option to record. Audio is saved on your Mac.
STEP 2Transcribe
Well, the meeting is next week, but please change it to Tuesday, no, Wednesday at 11 o'clock. Also, it would be helpful if you could share the materials first.
Whisper on your Mac turns what you say into text.
STEP 3Clean up
Um, about next week's meeting, Tuesday—no, WednesdayPlease change it to 11 o'clock.Also, Materials tooin advanceIt would be helpful if you could share it.
Rephrase the last part. Remove filler and correct punctuation.
STEP 4Insert
Please change next week's meeting to Wednesday at 11 o'clock.
It would also be helpful if you could share materials in advance.
Go to the input field where the cursor is. If you cannot find the input field, copy it.
Measured on October 5, 2026, MacBook Air M4, 32GB, Python 3.11. The voices are 24 examples of synthesized voices (including 4 examples of noise) by Kyoko and Reed of macOS. Transcription is Whisper large-v3-turbo, formatting is Qwen3.5 9B.
The series of processing consists of 14 cases, and the median value is silence detection → transcription → formatting. Transcription took 1.00 seconds and formatting took 1.01 seconds. It does not include the process of stopping recording, capturing the screen, or pasting text. Measurements were not made using the actual microphone and actual screen, and the results may not be the same for all utterances and environments.
The character error rate is the value obtained by dividing the number of character insertions, deletions, and replacements in the reference sentence by the number of reference characters. Punctuation marks, spaces, full-byte characters, numbers, and equivalent English notation are arranged according to fixed rules.
The numbers are verification results in a development environment and do not guarantee quality or speed. The demo text is a sample for illustrative purposes only.
Use AI without handing over your data.
Turn your voice into words. Turn characters into sentences that can be communicated. However, I want to decide where to send it myself. Orqnote is a Mac voice input app for those people.