Almost every voice recorder records properly. What sets them apart is what happens afterwards: who transcribes, where, and how much of your time it takes.
There are three families: the classic recorder, the smartphone and the AI recorder. The right choice depends less on budget than on what you record, and on what you must do with it next.
Three families of devices
| Family | What it does | What is left to do |
|---|---|---|
| Classic voice recorder | Records an audio file, offline. Models from Sony, Philips or Olympus are the typical example. | Transfer the file, transcribe it, summarise it. |
| Smartphone | Records with the built-in app, and often transcribes on recent models. | Format the transcript, handle calls and notifications while recording. |
| AI voice recorder | Records in one gesture, then a service transcribes and summarises. The result lands on your phone. | Review the summary. Decide whom you trust with the audio. |
Which recorder for which need
Answer the questions in order. The first "yes" gives you the family.
- You dictate alone, microphone close to your mouth?yes : Your smartphone, or an entry-level recorder.
- You must keep the original file intact: research, archive?yes : A classic recorder that records in WAV.
- You record lectures in a large hall?yes : A classic recorder with a directional microphone, in the front rows.
- You record meetings or appointments several times a week?yes : An AI recorder: it spares you the transcript and the minutes.
- None of the aboveyes : Your smartphone. Come back to this question if your use grows.
Reading a spec sheet
Spec sheets line up numbers that matter for music, rarely for speech. Here is what each line means, and what is enough to record a voice.
| Spec | What it means | Enough for speech |
|---|---|---|
| Microphones | Their number and pattern: an omnidirectional mic picks up the whole room, a directional one aims at a source. | Omnidirectional for a meeting table, directional for a distant speaker. |
| Format | MP3 compresses, WAV keeps everything. | MP3 day to day, WAV if the file must be archived or analysed. |
| Bit rate (kbps) | The amount of data per second in MP3. | 64 to 128 kbps. |
| Sample rate (kHz) | How many times per second the sound is measured. | 16 kHz is enough for speech recognition; 44.1 kHz serves music. |
| Battery | Recording hours on one charge. | Your busiest day, plus a margin. |
| Voice activation | The device only records while someone speaks. | Unnecessary, and it often clips the start of sentences. |
| Connector | How the file leaves the device. | USB-C, seen as a USB drive, no software to install. |
How many hours fit in the memory
File size depends on format and bit rate, not on the brand. For one hour of speech:
An 8 GB memory therefore holds about 140 hours of 128 kbps MP3, but barely 12 hours of stereo WAV. For speech, "CD quality" rarely doubles clarity and always multiplies size.
The criteria that actually matter
- Sound pickup, by distance. Dictation happens at twenty centimetres, a meeting at two metres. A device tuned for the first picks up the second poorly.
- The gesture to start. A physical button you find without looking beats a menu. A recording you forget to start is the only one truly lost.
- The way out. Can you get the audio in a standard format (MP3, WAV, M4A) and the transcript as text? A device whose data only leaves through its maker's app ties you to that maker.
- Transcription: included or not. If not, add the time or tool it takes, see How to transcribe audio to text.
- Where the data is processed. With an AI recorder, the audio goes to a server. Who runs it, in which country, under which law, and is the audio used to train models?
- The three-year cost, detailed below.
Working out the three-year cost
You pay for an AI recorder twice: at purchase, then every month if its free tier does not cover your use. Two fictional devices, to show the method:
The device that is cheaper to buy costs three times more to use. Before comparing prices, estimate your recording hours per month and check what each free tier actually covers.
The right choice for your use
| Use | What is enough | Why |
|---|---|---|
| Dictating letters or notes | Smartphone or classic recorder | Close voice, one person, the text gets reworked anyway. |
| Ideas on the go | Smartphone or AI recorder | What matters is the delay between the idea and the recording. |
| University lectures | Classic recorder with a directional microphone | The lecturer is far away, the hall echoes, the lecture lasts two hours. |
| Frequent meetings and client appointments | AI recorder | The time saved is the transcript and the minutes, several times a week. |
| Research interviews | Classic recorder in WAV, or AI if you control the processing | The original file must be kept and the transcript exact, see How to transcribe an interview. |
| Professions bound by confidentiality | A device where you know where the audio is processed, and by whom | Lawyers, notaries, healthcare: professional secrecy extends to what you hand to your providers. |
A recorder for lectures
A lecture hall is the hardest case: the lecturer speaks ten metres away, the room echoes, and two hundred people turn pages. Four tips:
- Sit in the front rows, facing the lecturer or near a PA speaker.
- Choose a device with a directional microphone, on a surface that does not vibrate.
- Plan battery for at least two lectures in a row.
- Ask the lecturer's permission. Many institutions have a written rule.
Good to know. A lecture is protected by copyright: French intellectual property law lists lectures among works of authorship. Recording one to revise is one thing; sharing it requires its author’s agreement.
A recorder for PC or Mac
To dictate at a computer, there is no need to buy a device. Windows 11 ships the Sound Recorder app, the Mac has Voice Memos, synced with the iPhone. A USB microphone on the desk clearly beats a laptop's built-in mic, which mostly picks up the fan and the keyboard.
Professions bound by secrecy
Lawyers, notaries, doctors, psychologists: your recordings hold information covered by professional secrecy. With a classic recorder, the audio stays in the device. With an AI recorder or a transcription service, ask these questions before using it:
- In which country is the audio processed and stored, and by which company?
- Which subcontractors can access it?
- Are the audio or transcript used to train models?
- How long are files kept, and how do you delete them?
Watch out. In France, hosting health data on behalf of a practitioner requires an HDS-certified host. If you record consultations, ask for the certification before any other criterion.
Hidden recorders: what the law says
"Spy recorder", "mini hidden recorder": the size of the device does not change the rule. In France, recording someone's words in a private setting without their consent carries one year in prison and a €45,000 fine (article 226-1 of the Criminal Code). The detail, and the exceptions, are in Recording a work meeting in France: what the law says.
AI recorders: Plaud, Notus and the rest
The AI recorder market has formed around a few devices. The best known is the Plaud Note, available now, in shops, covering up to 112 languages with a free tier of 300 minutes a month.
Notus makes other choices: processing and hosting in France with European models, recordings never used to train an AI, a free tier of 50 hours a month. It is a pre-order, with delivery estimated in the first half of 2027. If you need a recorder today, it is not the one. The Plaud Note or Notus comparison details three-year price, languages and data.
Traps to avoid
- The subscription you discover after buying. Check what the device does without a subscription, not only with one.
- Transcripts without export. Text you cannot copy or export cleanly does not save you time.
- WAV by default. Some devices ship set to maximum quality and fill their memory in a few hours.
- Recording without telling. The most discreet device changes nothing about the rule: people being recorded must know.
Frequently asked questions
Which voice recorder for university?
A classic recorder with a directional microphone, battery for at least two lectures, used from the front rows. A smartphone is fine for a small classroom, rarely for a lecture hall.
MP3 or WAV?
MP3 at 64 or 128 kbps for everyday speech: the file is ten times lighter and the transcript just as good. WAV for a recording to archive or analyse closely.
How much battery do I need?
Enough for your busiest day, plus a margin. A device that charges over USB-C charges anywhere.
Does an AI recorder work without internet?
Most record offline and transcribe once the audio has synced with the phone. Check for the model you are considering if you often record without a network.
Can I use my phone as a voice recorder?
Yes, and it is often enough. Settings and limits are covered in Voice recorder on iPhone and Android.
Key takeaways
- Choose by what you do with the audio, not by the spec sheet.
- For speech, 128 kbps MP3 and 16 kHz are plenty.
- For frequent meetings, an AI recorder saves you the minutes.
- Compare the three-year cost and where the data is processed.
- Discreet or not, a recording happens with the knowledge of those recorded.
Notus is an AI voice recorder designed in France: one button, then the transcript and summary on your phone in about thirty seconds.
Discover Notus →