I need to transcribe team meetings about software development. Otter keeps mangling our jargon and acronyms. Tried the custom vocabulary feature but the results are hit or miss.
What's the actual step-by-step to make this work? I'm on the Pro plan. Need specific examples for tech terms. How do you format the list? Does it handle case-sensitive stuff like "API" vs "Api"?
Otter's custom vocabulary is fundamentally broken for technical acronyms, and you're right to notice the hit-or-miss results. The feature is designed for proper nouns, not case-sensitive jargon.
Here's the only method that gave me consistent results:
- Format your list as one term per line, exactly as you want it transcribed. For "API", you must also include "A.P.I." on a separate line because the engine often expands acronyms phonetically. Your list should look like:
```
API
A.P.I.
Kubernetes
GraphQL
CI/CD
SaaS
IaC
```
The system does not handle case variations intelligently. If you only add "API", it will still sometimes output "Api" or "apee". You have to brute-force it by adding common mis-hearings.
Even then, it's unreliable for short terms spoken in fast conversation. It works better for longer project codenames. For engineering standups, I gave up and now run the raw transcript through a simple post-processing script that does a find-and-replace from a dictionary.
—davidr
That's a solid point about having to add "A.P.I." separately. I tried it with "CRM" and had the same issue - it kept coming out as "see are em" in the transcript unless I spelled it out letter by letter.
Does adding the phonetic version actually work for everyone? I still get "Api" sometimes even with both entries in my list. Makes me wonder if the order in the file matters, or if it's just a crapshoot.
Haha yeah, the phonetic brute-force method is hit-or-miss. I've had better luck with it for longer acronyms like "IPC" or "SLA" where the mis-hearing is more predictable. But for short ones like "CRM" or "API" it's never 100%.
Order in the file doesn't matter as far as I can tell - Otter seems to process the whole list as a bag of tokens. What _does_ matter (and this tripped me up) is whether you're uploading the list through the web app vs the voice commands. Web upload seems more stable. I've had the custom vocab setting just vanish after I saved it in the mobile app once.
Also worth noting: if you're in a noisy room or people talk fast, the engine just ignores your custom terms more often. It's like it falls back to its base model no matter what you feed it. Are you seeing it fail mostly in rapid-fire discussion sections or also in slower parts?
automate or die
Brute-forcing the list with phonetic entries isn't a solution, it's a workaround for a broken feature. You're paying a premium price for a tool that can't handle basic acronym transcription.
The real issue is Otter's engine isn't built for the precision technical fields require. Your find-and-replace script workaround proves that - you're paying for Pro and then doing the actual job yourself. Have you calculated the time cost of building and maintaining that script versus just using a different service?
trust but verify