1Scope
This method covers devices whose primary purpose is capturing or responding to speech: AI pendants and pins, voice recorders, and mic-equipped glasses, rings and earbuds. General-purpose smartwatches and fitness trackers are out of scope, though they may come up in the writing.
Scores are assigned from published specifications and cited reporting. Every device page lists those sources so a figure can be checked. The ranking is the weighted order of shipping devices; Hardware is the directory of the objects.
2Axes and weights
Seven axes, scored 0–10 in half-point steps. Weights are fixed and applied identically to every shipping device. Changing a weight re-orders the ranking; it does not quietly rescore devices that have not been re-read against the new weights.
| Axis | Weight | What it measures |
|---|---|---|
| Capture | 23% | Mic array, far-field pickup, wind and noise handling, and capture reliability. |
| Intelligence | 17% | Transcription, summaries, and whether a standalone assistant can act — connectors, tools, invoke from the device — not only a cloud recap of a recording. |
| Privacy | 16% | On-device processing, a visible recording indicator, and whether you can export and delete. |
| Battery | 14% | Real continuous-capture endurance and how much recharge friction it imposes. |
| Wearability | 12% | Weight, comfort across a full day, and whether you can invoke the system on the body — one hand, without fishing for a phone. |
| Look | 7% | How it reads in public as an object you would wear: materials, silhouette, jewellery versus gadget. A 10 is a piece you would forget is a recorder; a 3 is obvious electronics. Not the website. |
| Value | 11% | Hardware price and whether a usable free tier exists, measured against what you get. |
3Wearability and intelligence
Wearability is grams, comfort, and whether a full day reminds you the object is there — including whether you can invoke the system on the body: one hand, without fishing for a phone. A long press on a ring is wearability at the top of the scale, written in the note, not a fourth published crown. A MagSafe card that records from a button is a carry score. Clips that listen all day still send Actions to a phone; that lives on intelligence.
Intelligence is transcription, summaries, and whether a standalone assistant can act. A mature recap pipeline is still competent. An on-finger agent that turns a conversation into an action through connectors is a different kind of competence, written in the intelligence note. App-first summaries score lower on that axis than an agent you can hold. Capture remains the first weight: a ring that cannot hear a table does not launder that failure through a long press.
4The 0–10 scale
The same scale is used on every axis, so a 7 on battery and a 7 on look can be read the same way. Scores move in half points — 9.5, 8.5, 4.5 — not only integers. Intermediate scores interpolate: a 9.5 is a named reservation short of a 10; a 9 is closer to category-leading than a 7; a 5 is a named weakness that still does some of the job.
| Score | Meaning |
|---|---|
| 10 | Category-leading on the published record; no material reservation. |
| 7 | Competent, with a named reservation. |
| 4 | Does the job of the axis, with a clear weakness. |
| 1 | Fails the job of the axis. |
| 0 | The capability is absent. |
5How the total is calculated
Each axis is scored 0–10 in half-point steps. The total is the sum of each score multiplied by its weight, giving a figure out of 10. Sorting uses the full-precision value; the displayed figure is rounded to one decimal place. The total is never stored — it is recomputed from the axis scores every time the site is built, so a published number can never fall out of step with the scores behind it.
6Price and plan
The ranking desk prints hardware price and the plan label, not a year-one lump sum. Many devices ship with a usable free tier — 300 transcription minutes on Plaud Starter and Vocci Starter — and an optional paid SKU. Treating twelve months of Plus as the price of the object makes a $189 card look like a $288 product. The paid plan is what you pay after you outgrow the free minutes, not a cover charge to turn the device on.
Value still accounts for what a working year costs if you buy the encoded plan, and that figure remains the third tiebreaker. Device pages may print it as a footnote under Cost. It is not the ranking column.
7Which devices are ranked
A ranking you cannot buy from is not useful, so a device takes a position only once it is generally available. Announced, preorder and discontinued objects stay in Hardware and leave the ranking.
8Ties
Where two devices reach the same total, the order is decided in this sequence:
- Higher capture score.
- Then higher privacy score.
- Then lower year-one cost — purchase price plus twelve months of any required subscription.
- Then alphabetically by slug, so the result is always deterministic.
Totals are compared with a small tolerance. Two mathematically identical totals can differ in the last decimal place of floating-point arithmetic, and ordering devices by that difference would mean the chain above never actually ran.
9Unpublished specifications
A number the manufacturer does not publish, or publishes twice with irreconcilable values, is omitted rather than guessed. On the device page that absence is printed as Not published, which is itself a fact about the manufacturer. Optional spec fields exist for this reason: requiring a weight or a microphone count would force a figure from a teardown or a guess, and that is the one failure a scored ranking cannot survive.
10Corrections
Specifications are sourced and linked on every device page. If a figure here is wrong, it is a mistake rather than a position, and we would like to know — see About.