STT is the core technology that makes live captions possible. StreamTranslate uses Deepgram Nova-2 — the fastest and most accurate STT engine available — to deliver captions under 400ms.
Try STT Captions FreeSpeech-to-Text (STT) is the technology that converts spoken audio into written text automatically. You'll also see it called Automatic Speech Recognition (ASR) — these two terms describe the same underlying technology from slightly different angles. STT emphasizes the output (text), while ASR emphasizes the process (recognizing speech patterns).
STT systems work by analyzing audio waveforms and matching them against models trained on vast amounts of transcribed speech data. Modern STT uses deep neural networks that have been trained on millions of hours of audio, allowing them to recognize speech across different accents, speaking rates, microphone types, and acoustic environments with high accuracy.
For live streaming, STT needs to work in real time — meaning it processes audio as you speak rather than waiting for you to finish a sentence before beginning transcription. This "streaming STT" approach is what makes live captions possible, as opposed to the "batch STT" used for post-production transcription of recorded files.
The quality of your STT engine directly determines the quality of your live captions. A poor STT engine produces captions full of word substitutions, dropped words, and nonsensical phrases that misrepresent your speech and can be embarrassing or confusing for viewers. A high-quality STT engine like Deepgram Nova-2 produces clean, accurate captions that viewers can actually rely on to follow your content.
StreamTranslate uses Deepgram Nova-2 because it consistently outperforms competing STT engines on independent benchmarks, particularly on conversational and informal speech — exactly the type of audio typical in gaming and live streaming contexts. Nova-2 handles fast speech, gaming vocabulary, filler words, and overlapping audio better than older STT models.
For international streamers, STT quality also affects translation quality. StreamTranslate's NMT translation layer receives its input from the STT transcription, so errors in the STT phase produce errors in the translation. High STT accuracy means your multilingual captions in all 125+ supported languages are also more accurate.
The most accurate streaming STT engine available, with industry-leading benchmarks on conversational audio — the kind of speech found in gaming and live content.
Nova-2 returns transcribed text in 100-200ms from receiving audio, enabling sub-400ms total caption latency from speech to visible overlay.
Add gaming terms, character names, and streamer-specific vocabulary to improve STT accuracy for your specific content niche.
STT converts spoken audio into written text automatically. It powers live captioning, voice assistants, dictation software, and real-time translation tools for streamers.
STT and ASR are effectively synonymous. STT is the more consumer-facing term; ASR is more technical. Both describe the same technology of converting speech to text.
StreamTranslate uses Deepgram Nova-2, the industry's leading real-time STT engine with best-in-class accuracy on conversational and streaming audio.
Deepgram Nova-2 achieves over 99% accuracy on clean audio. Gaming streams with background audio typically achieve 92-97% accuracy depending on microphone quality.
Yes. Deepgram Nova-2 returns transcribed text in 100-200ms — fast enough for captions to appear synchronized with speech on live streams.