A product name can be spoken clearly while its spelling remains unexpected. Personal names, model labels and invented brand names therefore need more than a confident guess based on sound.
The key points
- A proper-name test needs the correct spelling independently of the generated subtitle.
- Compare each spoken occurrence with the untouched output and document the recording conditions.
- BestPost generates subtitles that you can correct word by word.
- This guide explains the procedure rather than reporting a measured recognition rate or promising error-free names.
This test separates recognition, spelling and correction. Names in the examples are fictional, and the article reports no collected BestPost recognition results.
Why does a name test need a separate reference?
A subtitle can sound plausible while showing the wrong name. You need to establish the intended spelling first. The generated output cannot serve as its own reference for what should have been written.
Use a reliable source from your work, such as spelling confirmed by the person or current product documentation. Also record which version of the name is actually spoken in the clip.
BestPost generates editable subtitles. This test examines names in your selected material. It does not claim that BestPost recognises every name or every language without errors.
Separate the spoken wording from editorial preference. If the speaker uses a short form, replacing it with the full official product name may be a deliberate edit. The expanded form is not automatically the only correct transcription of what was said.
Product names can keep their spelling
BestPost creates the subtitles. You can correct individual words before they become part of the picture.
- Creates subtitles, you fix every word, BestPost burns them in
- One style preset for the subtitles of every video
- Runs in the browser, on your computer and on your phone
How should you select test passages?
Collect name occurrences from typical videos. Describe whether the speech is clear, background music is present or several people are speaking. These conditions belong in the test description, not only in explanations added after problems appear.
Record the relevant details before testing:
- Establish the confirmed spelling of each name.
- Identify where the name is spoken.
- Describe the language and audible conditions.
- Mark unclear or interrupted pronunciation.
- Note any written name visible in the picture.
The record then shows what information the material contains. A clearly displayed name and an audio-only name are different starting conditions.
Do not select only striking errors if you want to describe ordinary work. You can deliberately examine difficult cases, but identify that selection as a challenge set rather than a representative collection.
What belongs in the review framework?
Decide which differences to distinguish. Incorrect capitalisation can have a different significance from substitution with an entirely different name. Keep both visible without redefining the categories to suit the results.
This table uses fictional names and outputs:
| Example reference | Example output | Review classification |
|---|---|---|
| Model Noria | Model Noria | Spelling matches |
| Model Noria | Model Gloria | Name replaced with another word |
| Northbow Pro | Northbow | Spoken component omitted, if Pro was audible |
| Leah Holm | Leah holm | Capitalisation differs |
| Noria S | Noria S or Noria C is unclear | Review the reference and audio passage again |
These are not observed BestPost outputs. They illustrate how you could describe your own results.
Do not force an ambiguous passage into an unquestioned right-or-wrong judgement. If human listeners cannot reliably distinguish the spoken letter, preserve that uncertainty. Product knowledge obtained later does not retroactively improve the recording’s clarity.
How do you run the test and correction?
Keep the untouched subtitle output before changing names. Otherwise, only the final wording remains and the original state cannot be demonstrated.
Apply the same process to each occurrence:
Establish the reference
Record the confirmed spelling and corresponding audio passage.
Preserve the output
Save the original subtitle before editing it.
Classify the difference
Review wording, spelling and audible clarity separately.
Inspect the correction
Check the corrected subtitle in the finished video.
The test and the editorial finishing remain distinguishable. A successful correction shows that the final version improved; it does not make the original automatic recognition correct retrospectively.
BestPost supports word-by-word correction and burning subtitles into the video. You maintain the separate test record yourself. This article promises neither an automatic name report nor a trainable recognition dictionary.
The clip gets a description too
BestPost writes from the footage. You can review names in the accompanying text before publication.
- Writes the description, YouTube title, hashtags and first comment from your video
- Every text stays editable until you post
- Its own text per platform, the same video file
How do you count results without a misleading rate?
Choose the unit beforehand. You can count distinct names or every spoken occurrence. If one name appears repeatedly, the two approaches answer different questions.
Make the counting method explicit:
- State the number of distinct names.
- State the number of occurrences actually reviewed.
- Separate clear cases from ambiguous audio passages.
- Describe the kinds of correction required.
- Identify the limits of the video and language selection.
An easily recognised repeated name then cannot silently stand in for a diverse collection. A high rate from a few clear examples also says little about unfamiliar recording conditions.
No general recognition accuracy is provided here. If calculating your own percentages, state the numerator, denominator and treatment of ambiguous cases. Otherwise, the number suggests more certainty than the test supports.
How can the findings improve later videos?
Recurring differences reveal which parts of your editorial review need attention. Keep a confirmed name list for the team and compare it with the actual footage.
When recordings are hard to understand, inspect the audio before publication. A visually attractive subtitle does not make an unclear source recording unambiguous. Styling supports reading; it does not replace correct wording.
After correcting a name, review the visible result. A long product label may change the line break. Checking spelling and checking readability are therefore separate parts of preparing the final post.
BestPost handles the initial subtitle creation. A documented reference check helps you protect the names by which people recognise the brand or product. That is the practical value of this test, without claiming universal error-free recognition.
For a closer look at the related task, read Review Automatic Captions Before Publishing. To check the next decision, see Measure editing effort for video descriptions.