Why short demos mislead
A short preview is chosen to flatter the voice. It rarely contains your channel's names, numbers, jargon, or the long explanatory sentences a real episode needs, and it is too short to reveal drift, a pace that tires the ear, or a pronunciation that is wrong every single time.
Use previews to build a shortlist, not to decide. Three to five voices in your language and style are enough to test properly.
Build a test script from a real episode
Take a script you have already published, or the next one you plan to, and make sure it contains the moments that break voices.
- Names of people, products, and places your channel mentions often.
- Numbers, dates, prices, and abbreviations.
- One long sentence with a parenthetical aside.
- A question, an exclamation, and a quiet, serious line.
- Your usual intro, sponsor read, and call to action.
Score every voice the same way
Generate the full test script with each shortlisted voice using the same settings, then listen the way a viewer would: start to finish, at normal speed, while doing something else for part of it. Write a score for each criterion before moving to the next voice.
- Pronunciation: are names and numbers right, and can you fix the wrong ones once with a pronunciation dictionary?
- Pacing: does it breathe at commas and paragraph breaks, or rush through?
- Consistency: does the last minute sound like the first?
- Listening fatigue: after ten minutes, would you keep listening?
- Edit cost: change one sentence and regenerate it; how long does the fix take, and does it blend in?
- Delivery package: do you get subtitles that match the final audio?
- Rights: does the plan you will use include commercial use for monetized videos?
Check what a season costs
Price a voice by the episode, not the month. A ten-minute episode is about 10,400 characters, which is 20,800 credits on NarrateHQ's Natural lane. That fits about 8 such episodes a month on Starter and about 31 on Creator, before any regenerations. More expressive lanes cost more per character, so test whether your shortlist actually needs one.
Leave room for re-takes. A voice that needs three attempts per section costs three times as much as one that lands first time, which is another reason to score edit cost honestly.
Lock the voice before episode one
Once a voice wins, treat its settings as part of the channel's identity. Keep the same voice, lane, rate, and pitch for every episode, keep the pronunciation dictionary up to date, and save a template script with your intro and outro so each new episode starts from the same place. Changing voices mid-series is noticeable to returning viewers, so it is worth getting the choice right before the first upload.