AI recorder marketing invents urgency; specs invent jargon. This glossary defines the 28 terms that actually recur while comparing devices — in plain English, with the evidence tier for any factual claim and a link wherever the concept has a whole page of its own.
7 terms
A–D
- Add-on minutes
- One-time transcription minute packs sold on top of a subscription — PLAUD's are 3,000 min/$59.99 and 6,000 min/$99.99, valid for 24 months from activation [SPEC]. Useful for a burst of heavy months without upgrading permanently; see our subscription comparison.
- AI recorder
- A pocketable hardware device that records audio and feeds it to speech-to-text AI for transcripts, summaries, and speaker labels. The recording half is offline and unmetered on every tracked device; the AI half is where subscriptions and cloud uploads live. How the whole pipeline fits together: how AI recorders work.
- Always-on capture
- Recording mode where the device listens continuously rather than waiting for a button press — Fieldy 3's model, and the pitch of the discontinued Limitless Pendant. Trades battery and social optics (and consent exposure — see our consent-law basics) for never missing an unplanned moment.
- ASR (automatic speech recognition)
- The software that converts recorded speech into text — the technology category under every transcript. “ASR model” and “STT model” mean the same thing; quality differences between devices trace to model size, training data, and how clean your microphone signal was.
- Bone conduction
- Capturing audio through physical vibration rather than air. Notta Memo's fifth microphone is a bone-conduction unit [SPEC]: pressed against a phone, it picks up call audio from the device body — a way to record both sides of a call on iPhones, where apps are blocked from doing so.
- Cloud transcription
- Uploading your audio to company servers where a large speech model transcribes it — the architecture behind PLAUD, TicNote, Notta, Pocket, and HiDock. Best English accuracy and AI features today, at the cost of audio leaving your device. Full trade-off analysis: local vs cloud.
- Diarization
- The software layer that segments a transcript by who spoke — clustering stretches of audio by voice characteristics. It's the “Speaker 1 / Speaker 2” machinery, and the most common accuracy complaint in owner reports. Plain deep-dive: speaker identification explained.
4 terms
E–L
- E ink
- A paper-like, ultra-low-power display technology familiar from e-readers. Most AI recorder screens are conventional instead — the PLAUD Note Pro uses an AMOLED panel [SPEC] — because full-color status and smooth updates matter more than weeks of screen battery in this category.
- Export formats
- The file types your transcripts can leave the app in — plain text, PDF, Markdown, SRT-style timestamped files, and app-to-app destinations like Notion. PLAUD's export guide confirms format choice with optional timestamps and speaker labels, though the exact extension list remains partially unverified in our research. Workflows compared in our exports guide.
- Free tier
- The transcription allowance included with hardware, no subscription — commonly 300 min/month (PLAUD, soundcore, SwitchBot MindClip). Mobvoi doubles it to 600; Notta offers 120; Fieldy 150; HiDock, Pocket, and Omi break the model with unlimited or local options. Whether a free tier covers you is a minutes-math question: try the calculator.
- LLM (large language model)
- The class of AI behind ChatGPT — and behind every “AI summary,” “action items,” and “Ask Plaud” feature in this category. The LLM never touches your audio; it processes the transcript that speech recognition already produced.
9 terms
M–O
- MagSafe
- Apple's magnetic attachment system on the back of iPhones. Several card-style recorders (PLAUD Note, WaveNote, Pocket) ship MagSafe-style cases so the recorder rides flat on your phone — where PLAUD's case adds its vibration-sensor call recording.
- MCP (Model Context Protocol)
- A standard that lets AI assistants like ChatGPT pull data from external services on your behalf. PLAUD ships an MCP server so ChatGPT can retrieve your PLAUD transcripts directly [SPEC] — an alternative to copy-paste export workflows.
- Meeting bot
- Software that joins Zoom/Teams/Meet as a visible participant and transcribes from inside the call — Otter's approach. Hardware AI recorders are not meeting bots: they capture room or speaker audio from outside the meeting software. The three capture methods compared: recording Zoom, Teams & Meet.
- MEMS mic
- Micro-Electro-Mechanical Systems microphone — the tiny, mass-produced, silicon-built microphone in phones and AI recorders. Multiple MEMS mics form an array whose geometry determines pickup range; PLAUD's NotePin S carries two [SPEC], and the Note Pro's upgraded array roughly doubles capture range over the base Note [SPEC].
- MFi (Made for iPhone)
- Apple's licensing program for authorized accessories. Anker markets the soundcore Work as an MFi-certified AI voice recorder for iPhones [SPEC] — a dependability signal for iOS pairing that uncertified rivals can't claim. More in our iPhone & Android guide.
- Offline recording
- Capturing and storing audio with no network connection — a capability of every tracked device. Do not confuse it with offline transcription: recording offline is universal; transcribing offline is rare (iFLYTEK, Omi's local mode).
- On-device transcription
- Transcription computed on the recorder or your phone, with audio never uploading. iFLYTEK's SR302 Pro does it fully offline in five languages [SPEC]; Omi runs unlimited transcription locally on your phone [SPEC]. Private and subscription-free, with English accuracy behind cloud leaders [REVIEW — mightygadget].
- One-party / all-party consent
- The two US consent frameworks for recording: one-party means any participant (like you) may record; all-party (“two-party”) requires everyone's consent, as in California under Penal Code §632. Which applies depends on state and situation — our consent-laws guide explains (educationally, not as legal advice).
- Open-source hardware
- Devices whose designs and software you can inspect, modify, or self-host. Omi is the tracked example — open-source apps and SDK with a tinkerer community [SPEC]. The practical privacy benefit: your transcription pipeline can survive the company.
8 terms
P–Z
- Per-file cap
- The maximum length of a single transcription job — separate from monthly minutes. PLAUD limits single files to 5 hours [SPEC]; Otter's free tier caps conversations at 30 minutes [REVIEW]. All-day conference recordings are where this bites.
- Pickup range / pattern
- How far from the microphone speech stays usefully captured, and the shape of that zone. Range beats model quality as an accuracy driver — distance, crosstalk, and reverb cause most transcription failures. Vibe Dot advertises a 16-ft listening range from five mics [SPEC]; the base PLAUD Note's mono mic favors close placement.
- Piezo / vibration sensor
- A component that turns physical vibration into signal. PLAUD's MagSafe case presses a vibration-based sensor against the phone body to capture both sides of calls [SPEC] — the category's signature trick for iPhone call recording. Details: phone-call recorders.
- Speaker labels
- The “Speaker 1 / Speaker 2” attributions in a diarized transcript — renameable to real names in every major app, with imperfect stickiness per owner reports [OWNER]. Often a paid-tier feature: Pocket reserves auto speaker-naming for its premium plan [SPEC].
- STT (speech-to-text)
- Synonym for ASR — speech-to-text is the task; the two abbreviations are interchangeable. When owners dismiss AI recorders as “just STT + LLM with a nice wrapper” (a verbatim Reddit phrasing), this is the STT they mean.
- TCO (total cost of ownership)
- Hardware price plus subscriptions over a realistic ownership period — we compute three years. It's the number that reveals a $159 card costing ~$879 over three years on an Unlimited plan, versus $0 extra for iFLYTEK. Our method: review methodology; the tool: cost calculator.
- Transcription minutes
- The metered allowance consumed when audio is transcribed — not when recorded. Recording is unlimited on every tracked device; minutes only tick when AI processes audio. This recording-vs-transcription confusion is the single most common billing question in owner forums.
- Wearable recorder
- A recorder designed to be worn — pinned, clipped, strapped, or ringed — so it's present for unplanned conversations. Trades bigger-card battery for hands-free presence within ~30 cm of your mouth. The form-factor trade-offs: card vs wearable.
FAQ
Frequently asked questions
Nothing — they're the same technology seen from three angles. ASR (automatic speech recognition) is the field, STT (speech-to-text) is the task, and 'transcription' is the product you experience. When a spec sheet says 'cloud transcription,' it means an ASR/STT model runs on the company's servers.
Only AI processing of audio — never recording. Every tracked device records unlimited audio locally; minutes tick only when audio is transcribed. That's why a 300-min/mo free tier is really '5 hours of AI per month,' and why a heavy lecture schedule (~40 hrs/mo) blows past it. Our minutes calculator does this math for your schedule.
Close cousins: 'on-device transcription' means the model runs locally (iFLYTEK's handheld, Omi's on-phone mode); 'offline recording' means capturing without a network — which every device does. A device can record offline all day and still upload everything for cloud transcription later.
'Free' — as in free tier. 'Free transcription included' can mean 300 metered cloud minutes (PLAUD, soundcore), 600 (TicNote), unlimited standard-accuracy cloud (Pocket — with a conflicting 300-min claim on one page we flag in our research), or unlimited on-device (iFLYTEK, Omi). Our no-subscription guide separates these meanings with three-year cost math.
It's our evidence-labeling system: [SPEC] = manufacturer's own pages, [REVIEW] = independent third-party coverage, [OWNER] = owner reports from Reddit/Trustpilot/Amazon. We don't claim hands-on testing, so we show you exactly what kind of source every claim rests on — defined fully in our editorial policy.