Switchboard
Conversational speech
Spontaneous telephone conversations. WER % results.
Audio intelligence
Use custom vocabulary to get product names, acronyms, and domain terms right in every transcript. Gladia matches how your terms sound against the transcript and corrects them to your exact spelling. Hosted in the EU or US, so your data stays where you need it.
Test custom vocabulary on recordings full of your own jargon. No expiry, no credit card.
6.4% WER on Earnings22, the highest speech-to-text accuracy of 8 providers tested.
In our benchmark, Solaria-3 had the lowest WER of the providers tested on real calls and financial audio. Your custom vocabulary only has to cover rare names and jargon, not everyday mistakes.
Update your term list whenever your product catalog changes, with no custom model to train, host, or retrain. Custom vocabulary runs in the same request as transcription, for both async and real-time.
Upload a file, pass a URL, or open a real-time WebSocket session.
Set "custom_vocabulary": true and add your terms. For hard cases, add how the word sounds and an intensity between 0.4 and 0.6.
Gladia compares the phonemes of each transcribed word with your terms and replaces sound-alike errors with your exact spelling.
Run a first transcription, check what the model outputs, then choose the tool that matches the error. Custom vocabulary fixes what the model mishears, and custom spelling fixes what it misspells.
| Criteria | Custom spelling | Custom vocabulary |
|---|---|---|
| Matches on | The exact text in the transcript | How words sound (phonemes) |
| Fixes | Recognizable but misspelled terms, like “data-science” for “Data Science” | Garbled or sound-alike errors, like “sell force” for “Salesforce” |
| You provide | Every variant the model outputs | The correct term, optional pronunciations, and an intensity |
| Use it when | The term is recognized but written wrong | The term is heard but not recognized |
Use cases
Make sure the words your product relies on, from client names to SKUs and show titles, come through exactly as written in every transcript, summary, and caption.
Spell client names, project codenames, and internal acronyms correctly in every summary and action item.
Learn more →Capture product names, plan tiers, and agent jargon correctly, so QA scores, search, and analytics count the right terms.
Learn more →Recognize brand names and SKUs in real time, so the agent acts on the right product instead of a sound-alike.
Learn more →Spell creator names and brand mentions correctly in captions and subtitles, so every name reads right on screen.
Learn more →Add five terms or fifty for the same rate. Custom vocabulary is included in every plan at no extra per-feature cost, so better transcription accuracy never shows up on your bill.
Flexible pay-as-you-go for moderate audio volumes. Get started immediately.
Async at $0.61/hr
Real-time at $0.75/hr
* 50€ in free credits
Lower unit pricing for fast-growing teams. Commit upfront to unlock savings.
Async as low as $0.20/hr
Real-time as low as $0.25/hr
* 67% less than Starter
Annual plan with custom models, fine-tuning, debundled pricing, and more.
Custom
Start free with 50€ in credits, or book a demo to test custom vocabulary on your own audio.
