We are having trouble with technical terms in our automated captions. Can you recommend speech-to-text services that allow for custom vocabulary training?
Rev AI : Provides a dedicated Custom Vocabulary API supporting both streaming and asynchronous transcription jobs, allowing you to submit thousands of domain-specific phrases per job.[](https://docs.rev.ai/api/custom-vocabulary)
Google Cloud Speech-to-Text : Uses Speech Adaptation , letting you pass phrase hints and boost values in your requests to bias the model toward specific technical vocabulary.[](https://www.gladia.io/blog/custom-vocabulary-stt-accuracy)
AWS Transcribe : Supports Custom Vocabularies , enabling you to upload a list of preferred spellings, acronyms, and technical terms so the transcription job prioritizes them.
If you want to narrow this down, let me know:
Are you looking for a developer API to build a pipeline, or an end-user captioning tool?
What programming language or environment are you integrating with?
I can help recommend the best fit for your workflow.
Yes. If your captions are failing mainly on technical jargon, product names, acronyms, and domain-specific terms, several speech-to-text providers offer vocabulary adaptation. The key distinction is that most don't literally retrain the model for you—they bias/prompt the existing model toward your vocabulary.
Provider
Customization
Good fit
Amazon Transcribe
Custom Vocabulary + Custom Language Models; supports pronunciation/display forms
Best if you have a substantial technical corpus and need deeper domain adaptation.
Deepgram
Keyterm Prompting for up to 100 important terms; enterprise customers can request custom models
Strong option for automated video/streaming transcription where jargon accuracy matters.
AssemblyAI
Custom vocabulary / keyterm prompting; up to 100 custom terms in its streaming implementation
Good developer-friendly API if you primarily need a curated jargon list.
Microsoft Azure Speech
Phrase lists for lightweight boosting; Custom Speech for actual model customization
Good if you want to move from simple vocabulary hints to training with your own text/audio.
Google Cloud Speech-to-Text
Phrase boosting/adaptation and custom model fine-tuning
Worth evaluating for large-scale captioning with substantial domain data.
What I'd choose for your use case
If you're building an automated video-captioning pipeline, I'd shortlist:
Deepgram — easiest first experiment with a list like Kubernetes, kubectl, PostgreSQL, gRPC, OAuth, WebAssembly...; its current Keyterm Prompting is specifically designed for important terminology and industry jargon.
Amazon Transcribe — stronger candidate if you have lots of domain-specific training text and want a custom language model rather than just boosting individual words.
Azure Custom Speech — particularly interesting if you have labeled audio/captions and actually want to adapt the model to your organization's terminology and speaking style.
AssemblyAI — attractive if you want a straightforward API and a relatively small set of critical terms.
One important point: "custom vocabulary" ≠ "training." If Kubernetes keeps becoming "cooper netties," vocabulary boosting may solve it. If the model consistently struggles with an entire technical domain, accents, context-dependent terminology, or specialized speech patterns, you should look for custom model/domain adaptation instead. AWS explicitly distinguishes these two approaches.
If you tell me which STT engine you're currently using (Whisper, Google, Azure, AWS, Deepgram, etc.) and roughly how many technical terms you have, I can narrow this to the best 2–3 options and compare pricing, API integration, and actual customization depth.
Top Speech Recognition APIs with Custom Vocabulary Support
Google Cloud Speech-to-Text
How it works: Offers Speech Adaptation and Custom Speech Models . For quick adjustments, you can supply inline Phrase Hints or reusable PhraseSets to boost the probability of specific domain terms. For heavy industry jargon, you can train a fully custom model by uploading domain-specific text corpora to fine-tune the base Conformer architecture.
Yes. Several speech recognition APIs support customization specifically to improve transcription accuracy for specialized industry jargon, technical terminology, acronyms, product names, and uncommon words.
The key distinction is that some APIs let you supply a list of vocabulary terms or phrase hints at transcription time, while others let you train or fine-tune a custom speech model using industry-specific text or audio. If your goal is to recognize terminology used in fields like medicine, law, engineering, finance, or manufacturing, both approaches can help.
Best speech recognition APIs for custom vocabulary
Handling technical terminology in automated captions requires speech-to-text (STT) services that support custom vocabulary, phrase hints, or domain-specific language modeling.
Top speech-to-text platforms featuring robust custom vocabulary support include:
AssemblyAI : Provides a custom vocabulary feature via its Speech-to-Text API. It allows you to specify target words, domain-specific terminology, and custom spellings to bias the model toward recognizing hard-to-capture jargon correctly.
Microsoft Azure AI Speech : Offers extensive custom speech capabilities. You can upload text data and domain-specific glossaries to train or adapt base models, significantly improving recognition accuracy for specialized technical, medical, or product vocabulary.
Google Cloud Speech-to-Text : Utilizes "Speech Adaptation" and "Phrase Hints". This lets you inject context, technical terms, proper nouns, and product names directly into the transcription request to guide the acoustic and language model.
Speechmatics : Features a Custom Dictionary option supporting specific words, phrases, and "sounds_like" pronunciation hints for unusual terms or product codes that fall outside standard phonetic dictionaries.
Rev AI : Offers an automated transcription and captioning API equipped with custom vocabulary tools to help increase recognition accuracy for brand names and industry-specific lexicons.
If you share details about your workflow, I can help you choose the best fit:
Are you looking for an API for developer integration or an end-to-end captioning application?
What specific technical domain or industry do your terms come from (e.g., medical, software, engineering)?
For automated captions that struggle with technical terminology, I’d prioritize speech-to-text services that support custom vocabularies, phrase boosting, or domain-specific language model training. These features help a recognizer handle specialized jargon, acronyms, product names, scientific terms, and uncommon proper nouns.
There’s an important distinction: some services let you supply a list of terms to improve recognition, while others let you train a model using technical documents or corrected transcripts. If your captions consistently mistranscribe industry-specific language, the second option may provide a bigger improvement.
Recommended speech-to-text services
Azure AI Speech
How it works: Features Custom Speech via the Speech Studio portal. It allows you to upload plain text or structured text datasets containing your specialized vocabulary, acronyms, and formatting rules. You can also perform acoustic adaptation by uploading domain-specific audio paired with transcripts.
How it works: Supports Custom Vocabularies and Custom Language Models . You can upload a list of specialized phrases along with SoundsLike or International Phonetic Alphabet (IPA) guidance to ensure words with unusual spellings or pronunciations are transcribed correctly.
How it works: Uses a feature called Word Boost (contextual biasing), allowing you to pass a list of custom key terms and boost their weight so the model heavily favors them during transcription without needing a lengthy separate training job.
Best for: Developers looking for a straightforward, high-accuracy API with easy keyword prompting.
Supports custom vocabularies for specific words, acronyms, pronunciation hints, and preferred display forms. It also offers custom language models trained on domain-specific text to learn how terminology is used in context.
Best for: healthcare, finance, scientific research, enterprise transcription.
Training: vocabulary lists or a larger text corpus for a custom language model.
Uses speech adaptation with phrase sets, custom classes, and boost values to favor specialized words over similar-sounding alternatives. Google also documents a separate custom-model training capability.
Best for: applications needing contextual phrase hints and configurable recognition.
Training: lightweight phrase hints; custom-model training is a separate, more involved option.
Provides Custom Speech models trained with domain-specific text, as well as phrase lists that can be applied without training a new model. Custom models can support real-time and batch transcription.
Best for: organizations already using Azure or needing organization-specific vocabulary.
Training: phrase lists for quick improvements; custom model training for broader domain adaptation.
Offers Keyterm Prompting for its Nova-3 and Flux models. You provide important terms or phrases in the API request, without needing to train a model. Keyterms can also be updated during supported streaming sessions.
Best for: real-time transcription, voice agents, call centers, and terminology that changes by customer or conversation.
Training: no separate training required for keyterm prompting.
Documentation:
5. Speechmatics
Provides a custom dictionary through its additional_vocab setting, including alternate pronunciations through sounds_like. It supports batch, real-time, and agent transcription.
Best for: specialized vocabulary, proper nouns, acronyms, and difficult-to-pronounce terms.
Training: add terms and pronunciations per transcription job.
Documentation:
Quick comparison
API
Add vocabulary without training
Train a custom model
Pronunciation support
Amazon Transcribe
Yes
Yes
Yes
Google Cloud
Yes
Yes, separate feature
Via adaptation and model capabilities
Azure AI Speech
Yes
Yes
Custom Speech supports broader adaptation
Deepgram
Yes
Keyterm prompting doesn't require training
Term prompting
Speechmatics
Yes
Not required for dictionary use
Yes, via sounds_like
The options above are documented by the respective providers; custom-model availability, languages, and feature limits vary by model and API version.
Which approach should you choose?
I would narrow the choice to three options depending on how specialized your terminology is.
A few dozen to a few hundred unusual terms: Start with Deepgram Keyterm Prompting, Google phrase hints, or an AWS custom vocabulary. These are relatively easy to integrate and don't require model training.
A large domain-specific vocabulary where context matters: Evaluate Amazon Transcribe Custom Language Models or Azure Custom Speech. These are better suited to terminology whose meaning depends on the surrounding sentence.
Unusual pronunciations or acronyms: Test Speechmatics custom dictionaries and AWS custom vocabularies, where pronunciation guidance is available.
One important caveat: adding a term does not guarantee correct recognition. Over-biasing toward a word can introduce false positives, particularly when jargon sounds like an ordinary word. Google explicitly documents this trade-off for phrase boosting.
For a fair evaluation, test each API on the same representative audio clips, compare word error rate, and separately measure accuracy on your industry-specific terms.
What industry are you targeting, and do you need real-time speech recognition or transcription of recorded audio? I can narrow this down to the best-fit APIs and their specific customization capabilities.
Deepgram's Nova-3 and Flux models support Keyterm Prompting, allowing you to supply up to 100 important terms or phrases, with a 500-token limit per request. It's designed for industry jargon, product names, acronyms, and technical phrases. It also supports streaming transcription, making it a strong candidate for live captions.
Best for: Live captioning systems where you want to improve recognition without training a new model.
Amazon offers both custom vocabularies and custom language models. Vocabularies help with specific words, pronunciations, acronyms, and preferred display forms. Language models learn contextual relationships from domain-specific text, helping distinguish terms that sound alike, such as “ice floe” and “ice flow.”
Best for: Technical training videos, scientific presentations, specialized lectures, or large-scale captioning with substantial domain-specific text data.
Google provides phrase hints, phrase boosting, and custom classes through its speech recognition adaptation features. You can emphasize technical terms or provide collections of related terms, such as product models, chemical names, or internal project terminology.
Best for: Applications that already use Google Cloud, or teams that need flexible vocabulary adaptation in an API-based captioning workflow.
Azure supports lightweight phrase lists for real-time transcription and also offers Custom Speech for training or adapting models using supported text, pronunciation data, and audio with human-labeled transcripts. Phrase lists don't require training, but they aren't supported by its batch transcription API.
Best for: Real-time captions in Microsoft-oriented environments, or organizations that want to move from simple term hints to a customized model.
Keyterm prompting rather than traditional vocabulary training
Yes
Amazon Transcribe
Custom vocabulary
Custom language model
Yes
Google Cloud
Phrase hints and custom classes
Model adaptation
Yes
Azure AI Speech
Phrase lists
Custom Speech
Yes
These are differences in customization approach, not a guarantee that one service will be more accurate for your audio. Feature availability can vary by model, language, and API.
My recommendation
I would start by testing Deepgram and Amazon Transcribe against your existing captions. Deepgram is attractive if you mainly need to correct a list of recurring terms; Amazon is particularly worth evaluating if you have corrected transcripts and technical documentation that could help train a more specialized model.
Whichever service you choose, run the same representative audio through each system, using the same glossary, and measure how often your technical terms are transcribed correctly. Also check punctuation, timestamps, latency, and speaker changes if those matter to your captions.
One question to narrow this down
Are your automated captions primarily for live speech (meetings, webinars, or events), or for pre-recorded audio and video? That distinction will help me narrow the options to the best fit for your workflow.
