Why Meeting Transcription Should Happen on Your Device

Automatic meeting notes are useful. Someone records the call, a transcript appears, and nobody has to scribble minutes while trying to follow the conversation.

The question most businesses skip is where the recording goes to become a transcript. With most transcription tools, the answer is someone else’s servers. For a weekly stand-up, that may not matter. For a disciplinary meeting, a pricing discussion or a conversation with a client’s solicitor, it matters a great deal.

What happens with cloud transcription

Cloud transcription services follow broadly the same pattern. The audio is captured on a laptop or phone, or by a bot that joins the call. It is uploaded to the provider’s servers. Their software converts speech to text, and often produces a summary as well. The audio, the transcript and the summary are then stored in the provider’s system so they can be viewed, searched and shared.

That is how most software works now, and it is not wrong in itself. It does mean a full record of the conversation, word for word, now exists outside the business, and several questions follow from that.

Who can access it? The provider’s systems hold the data, so the provider’s staff and any subcontractors it uses may be able to reach it under certain conditions. Most reputable providers restrict this tightly. The only way to know is to read the terms and the data processing agreement.

Where is it processed and stored? Some services process data in the UK or EU. Others use servers elsewhere. Under UK GDPR, sending personal data outside the UK brings extra obligations, and a meeting recording is almost always personal data because it contains people’s voices, names and opinions.

How long is it kept? Retention periods vary. Deleting a recording from the app does not always mean it is deleted from every backup straight away.

Is it used for anything else? Some services reserve the right to use customer data to improve their models, sometimes with an opt-out. Again, the terms are the only reliable source.

What happens if the provider is breached? A transcription service holds many businesses’ private conversations in one place. That makes it a more attractive target than any single small business.

Cloud transcription is not automatically unsafe. But the business has handed a copy of its most candid conversations to a third party and is relying on that party’s security, policies and honesty. That is a risk decision, and it should be made deliberately rather than by default.

Why meetings are a special case

Most business data has already been through some kind of filter by the time it is written down. Emails get reread before they are sent. Documents get drafted and reviewed.

Meetings are different. People think out loud. They speculate about a competitor, comment on a colleague’s performance, mention a client’s financial difficulties or float a redundancy plan that may never happen. A transcript captures all of it, including the parts nobody would ever put in writing.

That is what makes transcripts valuable, and it is also what makes them sensitive. The small business security guide covers the basics of keeping business data safe. Meeting transcripts deserve the same care as anything else holding confidential information, and arguably more.

What on-device transcription changes

On-device transcription does the speech-to-text work on the phone, tablet or computer that made the recording. The audio does not have to be uploaded to a transcription provider to become text.

That removes several of the questions above. There is no provider holding a copy of the audio for processing, no third-party retention policy for that step, and no question of the audio being used to train someone else’s model.

It does not remove every question. The recording still exists, and the business still needs to decide where it is stored, who can open it and when it gets deleted. The device needs to be protected properly, with a passcode and disk encryption. On-device processing cuts down the number of parties involved. Good security habits still apply.

The trade-offs

On-device transcription has costs, and it is better to know them up front.

Models need downloading. The software that turns speech into text has to live on the device. That means a download before the first transcription, and it can be a sizeable one. Doing it on a decent Wi-Fi connection before an important meeting avoids a frustrating wait.

Devices have limits. A phone or laptop has a fraction of the computing power of a data centre. Large cloud services can run bigger models and may handle difficult audio, heavy accents or crosstalk better. On-device accuracy is good enough for most meetings, but not every one.

Speaker identification is harder. Working out who said what is one of the more demanding parts of transcription, and on-device approaches have to make compromises, particularly with long recordings and similar-sounding voices.

Newer hardware and software are often required. On-device features tend to depend on recent operating systems, which can rule out older devices.

Sharing takes an extra step. Cloud services are built around sharing a link with the team. With on-device transcription, sharing is more deliberate: exporting a file or putting it somewhere others can reach. For confidential meetings, that friction is arguably a benefit.

How Phonic handles it

Phonic is our first app. It records dictation and meetings on iPhone, iPad and Mac, with an Apple Watch companion. It was built around keeping transcription on the device, and the details are set out in the Phonic privacy policy.

Phonic transcribes recordings using Apple’s speech models running on the device. With Phonic Pro, speaker identification also runs on the device. Phonic does not ask you to create an account with us.

Each recording is saved as a .phonic package containing the audio, its metadata, the transcript and, with Phonic Pro, its Insights. These packages sit in your iCloud Drive, under your Apple Account and Apple’s terms, and you can see them in Files and Finder. If you do not use iCloud, recordings stay in a Phonic Library on the device. Either way, they are files you control, and you can move, export or delete them at any time. The storage help article explains how this works across devices.

Worth being clear about: if you use iCloud, the recording does leave the device, to your own iCloud Drive. The difference is that it is stored in an account you already control, rather than handed to a transcription provider to process and keep.

The first transcription in a language downloads Apple’s speech model for that language. With Phonic Pro, the first speaker identification downloads FluidAudio’s speaker models, which are about 120 MB. Both are reused afterwards. The models help article shows how to download them in advance, which is worth doing on Wi-Fi before your first long meeting.

With Phonic Pro, the Insights tab produces a summary, key points, action items, decisions, open questions, topics and talk time for each recording. It uses Apple Intelligence when it is available on your device, with a built-in fallback when it is not.

There are limits. Speaker identification uses the Sortformer model for recordings up to an hour. Longer recordings use clustering alone, which can merge similar voices. Phonic needs iOS 27, iPadOS 27, macOS 27 or watchOS 27, so older devices are out.

Phonic is free to download. Phonic Pro is a one-off £2.99 in-app purchase that adds speaker identification, system audio recording, Meeting mode and Insights.

Before you record anyone

Whichever tool you use, recording a meeting means processing other people’s personal data. Tell participants that the meeting is being recorded and why, keep recordings only as long as they are needed, and make sure the people who can open them are the people who should. The ICO publishes guidance on UK GDPR that covers the principles. This is not legal advice, and anything unusual is worth checking properly.

The practical takeaway

Cloud transcription is convenient and often very good. It also means a word-for-word record of your conversations sits with a third party. For routine meetings, that may be an acceptable trade. For confidential ones, keeping transcription on the device removes a whole category of risk, at the cost of a model download, some accuracy limits and the need for a recent device.

If you want help working out which of your business’s conversations and data need this level of care, get in touch.