---
title: "Comment by DemoVexa on \"Speech-to-text can't tell you if an AI voice said a name right\" | Hashnode"
description: "The point that a transcriber writes the likely spelling, so a near miss on a name simply vanishes, is the clearest explanation of this problem I've read. We run"
url: "https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33/comment/6af3d593108fc59f9c496b9d"
---

[Discussion](https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33)

Comment [Speech-to-text can't tell you if an AI voice said a name right](https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33)

[DemoVexa](https://hashnode.com/@demovexa)

[7h ago](https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33/comment/6af3d593108fc59f9c496b9d)

The point that a transcriber writes the likely spelling, so a near miss on a name simply vanishes, is the clearest explanation of this problem I've read. We run into the same thing with product voiceovers, where the risky words are acronyms more often than surnames. Whether "SQL" is read as letters or as "sequel" is a choice each product makes, and a transcript shows the same three letters either way. Keeping the researched pronunciation and the engine respelling side by side in one lexicon file is the habit I'd push hardest. It turns each fix into something the next video inherits instead of a one off correction. Being upfront that the 32 takes also set the thresholds is a nice touch.

[Logit Solutions AI](https://hashnode.com/@logitsolutionsai)

[7h ago](https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33/comment/6af3d593108fc59f9c496b9d#reply-6af3d7d0108fc59f9c497446)

Thanks! Acronyms are the same trap, you're right. Whichever reading a product picks, the lexicon entry records it as the target, the respelling makes the voice say it, and the sound check flags any take that says it the other way. We respell every acronym in our own videos for exactly this reason. And yes, the next video picking up the fix for free is the main reason the lexicon exists.

[DemoVexa](https://hashnode.com/@demovexa)

[5h ago](https://hashnode.com/posts/speech-to-text-can-t-tell-you-if-an-ai-voice-said-a-name-right/6af3a86b8e08d0169e437c33/comment/6af3d593108fc59f9c496b9d#reply-6af3f40b489220fa7f31eed9)

Respelling every acronym by default is a good rule. It saves arguing each one case by case, and the sound check catches the takes that slip. One thing I keep running into: some product teams say their own acronym both ways. When that happens, do you go with the reading from their docs and videos, or ask them which one they want?