Aiko

Click the feedback button in the app or send it here.

The v3 model is worse than v2 in too many cases. I tried releasing v3, but got a lot of emails about the quality being worse, so I ended up reverting it.

Not currently. It cannot be the default model as it’s worse than v2, so it would need support for downloading models, which is a huge amount of work.

Even the latest iPhone is not powerful enough to run the large model. It can maybe be done when the Whisper Distilled project supports multiple languages.

Not currently. Compared to Whisper, the SpeechTranscriber API:

• Supports far fewer languages
• Has no automatic language detection
• Has lower quality

The app tells the system not to turn off the screen. However, if you have enabled “low power mode”, the system may do so anyway. Try turning it off.

I don’t plan to support any editing. Export the transcription and edit it in a proper text editor.

The built-in transcription in iOS 26 and macOS 26 is fast and works well for the languages it supports. If it does what you need, use it. Aiko gives you more:

• Whisper supports about 100 languages.
• Transcribe any audio or video file.
• Export to text, text with timestamps, subtitles (SRT and WebVTT), JSON, and CSV.

• Aiko is a one-time purchase that works on Mac, iPhone, iPad, and Apple Vision Pro.
• Aiko is simple and only transcribes on your device.
• MacWhisper has more features, like speaker labels, more AI models, and optional cloud services.

Try both (see the Aiko trial) and see for yourself.

The app uses the OpenAI Whisper model and I have no control over the quality of its output. You could provide feedback about the problem in the Whisper discussions.

For names and technical terms, add the correct spelling in “Word replacements” in the settings.

I have no control over the supported languages. You could try to request it in the Whisper discussions.

This is unfortunately a flaw in the Whisper AI model and out of my control. This is usually caused by the audio not being clean (for example, a lot of background noise).

Some things you could try:

• Restart your device.
• Try enabling the “Skip silent parts” setting.
• Try enabling the “Reduce repetitions” setting.
• Try enabling the “Reduce repetitions even more” setting.
• If you don’t need timestamps, you can disable them being produced, which can sometimes reduce repetitions. Triple-tap on the “Translate to English” text in the settings and then disable the “Produce timestamps” setting.

Aiko divides the transcription text by sentences. Divide it into paragraphs.

Aiko shows one transcription at a time. A new transcription replaces the previous one, so export or copy a transcription you want to keep. The last transcription is shown again when you open the app (the “Restore last transcription at launch” setting).

Start a new transcription with the “+” button in the toolbar. Your recordings are kept, so you can transcribe an old recording again.

This is unfortunately a flaw in the Whisper model. Workaround.

This is unfortunately a flaw in the Whisper model. It learned from videos with subtitles, so in silent parts or at the end, it can add text like “Thanks for watching!”, Untertitel der Amara.org-Community, Sous-titres réalisés par la communauté d’Amara.org, or Teksting av …, or repeat “you” or “…” again and again. These texts have no meaning, and there is not much I can do about this.

Workaround: Try enabling the “Skip silent parts” setting.

Choose the language of the audio in the “Audio language” setting, instead of “Auto-detect”. Automatic detection can pick a similar language, like Welsh for English.

Ensure the “Translate to English” setting is disabled.

The language used in the “Prompt” setting may also affect the transcription language.

Not reliably. The Whisper AI model used by the app detects one language from the start of the audio and transcribes everything as that language, so parts in another language may come out translated. Writing a sentence that mixes the languages in the “Prompt” setting can sometimes help.

The Whisper AI model used by the app does not differentiate between Traditional Chinese and Simplified Chinese, so the result could unfortunately end up with either. Learn more in the Whisper discussion.

Try writing a sentence in Traditional Chinese or Simplified Chinese in the “Prompt” setting in the app to steer the model into using the right one.

iOS apps are fundamentally restricted from operating in the background for extended periods. This ironically even affects Apple’s official apps.

Any audio and video format that macOS and iOS supports. For example: .m4a, .wav, .mp3, .mp4, .mov. It does not support .ogg.

No, the AI model processes either transcription or translation, not both simultaneously. You need to run the process twice: once for the original transcription, and again with the “Translate to English” setting enabled.

The audio recordings can be found in the “Aiko” folder in the Files app.

Recordings are not automatically deleted after transcription.

macOS: Click “File › Show Recent Recordings” in the menu bar to open the recordings folder in Finder. Delete them from there.

iOS: Open the Files app and navigate to the “Aiko” folder.

You can also enable “Auto-delete recordings older than 7 days” in settings.

The error comes from iOS and means that it could not read the audio file. Sometimes converting the audio file to a different format, like MP3, resolves the issue.

macOS: Drag and drop the memo into the Aiko window. Note that because of a macOS bug, this can sometimes crash Aiko. If this happens, try sharing the memo from the Voice Memos app to Aiko instead.

iOS: In the Voice Memos app, tap the memo, tap the … button, tap Share, and choose Aiko in the app list.

iOS 26: Sharing from Voice Memos is broken due to an iOS bug. See workaround.

This is caused by an iOS 26 bug outside of my control. As a workaround, you can use a shortcut that appears in the Voice Memos share sheet:

1. Install the “Transcribe with Aiko” shortcut.
2. In Voice Memos, share the recording.
3. Look for Transcribe with Aiko in the share sheet. If you don’t see it, scroll to the end of the app row and tap Edit Actions.
4. If it’s still not listed, open the Shortcuts app, tap the Transcribe with Aiko shortcut, tap the ⓘ button, and enable “Show in Share Sheet”.

On an Intel Mac, transcription is very slow, as the model is optimized for Apple silicon.

Several factors can affect the transcription speed, including the performance of your device and the amount of available memory and CPU. Try closing down other apps or restarting your device before transcribing.

In the settings, make sure “Reduce resource usage” is off and “Use flash attention” is on.

The app includes the Whisper AI models, which support 100 languages. The download is 2.8 GB on Mac and 1.8 GB on iPhone and iPad.

On Mac, it uses the large model, which needs an Apple silicon Mac with at least 16 GB of RAM. On iPhone and iPad, it uses the medium model, or the small model when there is less free memory.

The role of the operating system (iOS/macOS) is to effectively manage the system’s resources and to safeguard against overheating. Apps are designed to utilize as many resources as they need to function optimally. In the event that resource consumption reaches a point that threatens to overheat the device, it’s the operating system’s job to limit such usage automatically.

If your device is experiencing overheating while using the app, it’s important to understand that the issue likely originates either from the operating system’s inability to manage resources effectively, or from an underlying hardware problem. In either case, the app itself is not responsible for the overheating.

To make the app use less CPU and battery, turn on “Reduce resource usage” in the settings. Transcription is then slower.

This is unfortunately not possible. The model has all the languages stored together in a way that makes it impossible to remove just some languages.

Not currently. Record the audio first, and then transcribe the recording.

This is called Diarization. It’s something I would like to support, but it has to first be implemented in the library the app uses.

The app does not support live transcription, but you could record the Zoom meeting, and after the meeting is finished, drop the recording into the Aiko window to transcribe.

Drag and drop the voice note into Aiko.

How to drag and drop on iPhone.

Telegram voice notes are stored in the format Ogg, which macOS/iOS cannot handle.

Workaround for iOS:

1. Download this app.
2. In Telegram, share the voice note to “Audio Converter”.
3. Select “AAC” as output format and tap the convert button.
4. Tap the share button and then choose Aiko.

Workaround for macOS:

1. Download this app
2. In Telegram, right-click the voice note and save it.
3. Open the saved voice note with the “Audio Converter” app.
4. Select “AAC” as output format and tap the convert button.
5. Save the converted file and open it with Aiko.

I would also recommend sending feedback to Telegram that they should support M4A for voice notes.

When the transcription is done, click the save button in the toolbar, and choose “SRT”.

For shorter subtitle lines, turn on “Optimize for subtitles” in the settings before you transcribe.

Download the audio using a service like dirpy or the macOS app Downie and then open the file in Aiko.

This cannot be supported built-in as downloading videos from YouTube is against their Terms of Service and Apple would likely reject such a feature.

The translation support is built into the AI model and it only supports translating to English. You could translate the result with Apple’s built-in translation (select the text, right-click, and choose “Translate”), or paste it into ChatGPT or Google Translate, but then the text leaves your device.

Aiko does not have a custom URL scheme, but you can trigger Aiko from the Shortcuts app, and the Shortcuts app does have a custom URL scheme.

Flash Attention is a special technique that helps the app process audio more efficiently while using less memory on your device. Think of it as a smarter way to handle the heavy lifting of transcription. It’s a setting in case it causes any problem. It should not though.

You can do it manually for free. See this guide.

You can also use the Replacicon app (paid).

iOS: There is a setting for it in the app.

macOS: Unlike iOS, macOS does not have the ability to have alternative icons. See the above FAQ. Here’s the old app icon.

IPA (International Phonetic Alphabet): [a.i.ko̞]

English approximation: ah-ee-ko

Yes, it’s native and written in Swift and SwiftUI.

On macOS:

1. Open Finder.
2. Click Finder in the menu bar, then Settings.
3. Go to the General tab.
4. Uncheck “Suggest file names”.

I don’t plan to localize the app.