Home/Methodology

Our review standard

Real Japanese audio. The same test. Every important error shown.

We do not rank transcription tools from feature lists or vendor accuracy claims. A native Japanese reviewer records the source, gives each service the same file, and checks the output against the intended meaning line by line.

Recorded in JapanNative-reviewedUpdated September 3, 2026

How one recording becomes publishable evidence

  1. 01
    Write a focused Japanese script

    Each recording isolates a real risk: homophones, names, corrections, speakers, noise, or business identifiers.

  2. 02
    Record the original audio in Japan

    We keep the source file and written reference so every output can be checked against the same intended sentence.

  3. 03
    Upload the exact same file

    Tools receive the same audio whenever their plans and technical limits allow it. Missing coverage is disclosed, never guessed.

  4. 04
    Review every material detail

    A native speaker checks wording, kanji, numbers, names, codes, corrections, and speaker labels—not just whether the paragraph looks fluent.

  5. 05
    Publish strengths, failures, and limits

    We separate meaning-changing errors from harmless formatting and state what the sample cannot prove.

Five checks used across every product review

We show raw counts beside a plain-language status. We do not combine unrelated observations into a universal “98% accurate” claim or an arbitrary star rating.

Test completion

Strong
Every assigned file produced a usable transcript.
Mixed
Most files completed, with a clearly disclosed coverage gap.
Concern
A repeated processing failure affected the evidence.

Japanese context

Strong
Almost every targeted homophone used the intended kanji.
Mixed
The transcript was readable but several contextual kanji were wrong.
Concern
Meaning-changing contextual errors were frequent.

Spoken corrections

Strong
All final corrected details remained clear.
Mixed
The correction survived, but at least one surrounding detail changed.
Concern
A final instruction, value, or responsibility changed.

Speaker handling

Strong
Speaker turns stayed correctly separated in the tested conditions.
Mixed
Useful separation with a limited error or an untested condition.
Concern
Speaker separation failed or obscured who said what.

Business identifiers

Strong
The intended letters and digits were preserved.
Mixed
Letters and digits survived, but punctuation or formatting changed.
Concern
A code, price, quantity, date, or other operational value changed.

“Incomplete” means the relevant test could not be run or finished. It is a coverage label, not an accuracy penalty.

A polished sentence can still contain a major error

FORMAT-ONLY DIFFERENCES

  • Optional punctuation or spacing changes
  • Arabic numerals written as equivalent kanji
  • Hyphens removed while every letter and digit remains correct
  • A defensible spelling variation that preserves identity and meaning

MEANING-CHANGING ERRORS

  • The wrong contextual kanji changes the intended action
  • A name, code, price, quantity, date, or destination changes
  • A spoken correction is lost or reversed
  • Speaker labels make responsibility unclear

Affiliate compensation does not change the test

We may earn a commission from clearly labeled product links. The source files, pass criteria, ranking, and published weaknesses stay the same whether or not a company offers an affiliate program. Normal links are labeled too.

Plans, interfaces, and models can change. Every review includes a checked date, and we update conclusions when new matched recordings reveal a different result.

Read our full affiliate disclosure.

Use the evidence

Choose the tool for the details your workflow cannot afford to lose.

Start with the matched comparison, then read the individual review for the workflow, pricing, and exact mistakes behind the recommendation.

See the current comparison →

Published September 3, 2026