Customising transcription

Registering technical terms and company names, default speaker names, and transcript display settings.

In meetings full of company names, department names, product names and industry phrasing, registering the terms in advance improves the chance of them being transcribed correctly. You register them under Settings > Preferences > Language. There are two kinds: your own terms, which apply to your meetings, and shared terms, which apply across the workspace.

Registering helps most for words you can see would otherwise be misheard. Registering large numbers of ordinary Japanese words, or normal dictionary words, barely changes accuracy.

There is no need to register your own company name or the names of people attending the meeting — those come from the meeting's own information. Names of other companies mentioned in conversation, contacts at client companies, and external product names, on the other hand, are worth registering.

What is worth registering for Japanese meetings

In Japanese speech, the common failure is the right sound written with the wrong characters. Registering words like these in advance means fewer corrections when you read the transcript back.

Type Example to register Typical error without it
Department names 品質保証部 the same reading written with different characters, such as “品質補償部”
Personal names (external, executives) 山田, 齋藤, 渡邊 replaced by a different surname with the same reading; old-form characters modernised
Names of internal documents and forms QC工程表, 日程遵守率 inconsistent spellings such as “QC工程票” or “日程順守率”
Product and project names project names, code names transcribed as ordinary nouns
Industry terms and abbreviations abbreviations used day to day on the ground turned into a different, similar-sounding word
Internal terms mixing English names that mix Japanese and English word boundaries fall differently from what you expect

Register proper nouns in the form actually spoken in meetings. If documents say “株式会社〇〇” but people say “〇〇さん” in conversation, register the spoken form.

In meetings that mix Japanese and English, the language setting itself affects the result too. See also Multi-language support.

Registering your own terms

Your own terms apply only to your meetings.

  1. Settings > Preferences > Language.
  2. Your terms field, type the word you want to register.
  3. Enter to confirm.
FigureScreenshot: your own terms under Settings > Preferences > Language

Registered terms shown as tags below the input field.

You can register up to 30. Because the number is limited, the efficient approach is to register words in the order they actually cause errors.

If shared terms have already been registered at workspace level, you can see them by opening “View terms applied to everyone in this workspace” below your own list. Those apply automatically, so there is no need to register them again yourself.

Registering terms for the whole workspace

Workspace terms apply to the transcripts of every member of that workspace. Words that should come out the same way no matter who in the organisation records the meeting — department names, your own product names — belong here.

Only workspace administrators can add and remove them.Settings > Workspace > General and edit them under Language. Members can view the list but not edit it. Up to 50 can be registered.

FigureScreenshot: shared terms under Settings > Workspace > General

The list of terms an administrator can edit, with the field for adding more.

Shared terms are visible to every member

Any member of the workspace can see the list of workspace terms. Registering a project name or code name shared only within a limited group, or the name of an individual at a client company, can therefore make that fact known more widely than intended. Choose what you register on the basis that it will be seen.

Examples

  • Registering your own service name spelled “Blokk” makes it less likely to be transcribed as “block”.
  • Registering a project name spelled like “Projekt DolFin” makes it less likely to be transcribed as “project dolphin”.
  • Registering a company name such as “Tekto” makes it less likely to be broken into similar-sounding words.

The same thinking applies in Japanese. If the same word comes out wrong in the same way meeting after meeting, that word is a candidate for registration.

What registering does not do

  • Word replacement: registering makes a word more likely to be recognised when it is spoken. It is not a feature that mechanically replaces one word in the transcript with another. A similar-sounding word can still survive.
  • Blocks of text: each term is passed to transcription individually. Registering a long sentence or paragraph changes nothing unless that sentence is spoken exactly.
  • Expanding abbreviations: abbreviations are transcribed as spoken. They are not expanded to the full name in the transcript. To check what an abbreviation means, asking chat during or after the meeting is quicker.

When it takes effect

Registered terms take effect from the transcripts of meetings recorded afterwards. Transcripts already recorded are never rewritten retrospectively. Where a past meeting has many errors, edit the transcript directly or check the content through chat.

Speaker names and transcript display

The speaker names shown in a transcript are decided by a different mechanism from term registration. By default, speech captured from system audio and speech captured from your microphone are only distinguished by colour, with no names attached. Attaching the meeting tool's display names to speech requires setup such as a browser extension. Which meeting tools are supported, and the steps, are in Speaker labels.

On-screen actions during a meeting — showing and hiding the transcript panel, pausing and resuming, copying and searching — are described in How transcription works.

If something is not working

If recognition does not improve after registering terms, the underlying audio may not be good enough. Check the microphone selection, system audio permission, and the distance between microphone and speaker along with background noise first. The steps are in Transcription problems.

Related articles

← Writing effective recipesMulti-language support →