Universal-3 Pro Technical Overview

2026-02-03 · Source: AssemblyAI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, medium

Summary

AssemblyAI has released Universal-3 Pro, a new speech-to-text model that introduces text prompt input for customized transcription outputs. This model allows users to influence transcription style, formatting, entity accuracy, speaker attribution, audio event tags, and code-switching through prompts. An out-of-the-box comparison with Universal-2, AssemblyAI's previous production model, demonstrated Universal-3 Pro's immediate improvements in correcting broken words, capitalizing proper nouns, and fixing sentence meanings without any prompting. Further demonstrations showed how specific prompts could accurately transcribe disfluencies like stutters ("it may") and add verbatim hesitations such as "ums" into the transcript, offering granular control over the output for various use cases. Users can access Universal-3 Pro via the `speech_models` parameter and experiment with the `prompt` parameter in their requests.

Key takeaway

For AI Engineers and NLP Engineers building transcription services, Universal-3 Pro's prompting capabilities offer unprecedented control over output customization. You should explore integrating the `prompt` parameter to fine-tune transcripts for specific domain requirements, such as verbatim capture of hesitations or precise entity recognition, significantly enhancing the utility and accuracy of your applications.

Key insights

Universal-3 Pro offers customizable speech-to-text transcription via text prompts, enhancing accuracy and output control.

Principles

• Prompting customizes transcription output.
• Model improves out-of-the-box accuracy.
• Prompts control disfluencies and formatting.

Method

Users provide audio files and a text prompt to Universal-3 Pro, which then generates a customized transcription. The prompt guides the model on aspects like style, entity recognition, and verbatim output.

In practice

• Use prompts for specific formatting needs.
• Improve entity accuracy with contextual clues.
• Add speaker attribution via prompting.

Topics

• Speech-to-Text
• Prompt Engineering
• Customized Transcription
• Universal-3 Pro
• Disfluency Detection

Best for: AI Engineer, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Prompt Engineering Workshop: Universal-3 Pro — Universal 3 Pro offers customizable speech-to-text via natural language prompts, enhancing transcription accuracy and control.
• Universal-3 Pro transcribes ASMR — Universal-3 Pro accurately transcribes whispering and non-speech audio events using advanced prompting capabilities.
• Gemini 3.1 Flash TTS — Gemini 3.1 Flash TTS offers highly granular, prompt-driven control over speech generation, including accents and vocal styles.
• SoundWise Launches Free Forever AI Audio and Video Transcription Tool for Unlimited Speech to Text Conversion — SoundWise offers a dual-engine AI transcription service, providing free, private, local processing alongside a faster, professional cloud-based option.
• How I Improved Speech-to-Text Accuracy — A two-pass LLM post-processing method effectively corrects common speech-to-text transcription errors.
• Nothing introduces an AI-powered dictation tool — Essential Voice offers AI-powered dictation with filler word removal, custom shortcuts, and multi-language translation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AssemblyAI.