Blurt

Product Introduction

1. Definition: Blurt is a free, open-source dictation application for macOS that provides real-time, high-accuracy speech-to-text transcription. Technically, it is a native desktop client that leverages a cloud-based AI transcription API.
2. Core Value Proposition: Blurt exists to provide a fast, accurate, and universally compatible voice typing tool for Mac users, enabling hands-free text input in any application with a text field. Its primary value is frictionless, context-aware dictation powered by a state-of-the-art speech recognition engine.

Main Features

1. Hotkey-Activated Universal Dictation: Users can dictate text into any focused application by simply holding down a configurable hotkey. This feature bypasses the limitations of built-in system dictation, working seamlessly in development environments, terminals, and communication apps like Slack and Discord.
2. AI-Powered Text Cleanup and Style Control: Beyond verbatim transcription, Blurt utilizes integrated AI to automatically remove filler words (e.g., "um," "uh") and can reformat speech into different writing styles (e.g., Formal, Casual) based on user-defined prompts. This is powered by the AssemblyAI Dictation API's real-time language model (LLM) processing.
3. Low-Latency, Cloud-Based Transcription Engine: The app streams audio directly to AssemblyAI's Dictation API, which returns initial transcription results in approximately 134 milliseconds (p50 latency). This cloud-first architecture eliminates the need for local model downloads and provides immediate access to the latest, most accurate speech model without app updates.
4. Multilingual and Code-Switching Support: The underlying Universal speech model supports dictation in 19 languages and can dynamically handle code-switching (changing languages mid-sentence) without requiring manual mode switching, making it ideal for multilingual professionals and developers.
5. Transparent, Open-Source Codebase: Released under the MIT license, Blurt's entire codebase is publicly available on GitHub. This allows for complete security auditing, customization, and serves as a production-ready reference implementation for developers building their own dictation applications.

Problems Solved

1. Pain Point: Slow, inaccurate, and application-restricted dictation tools. Native macOS dictation can be slow, inconsistent, and fails in many developer tools and web applications.
2. Target Audience: The primary user personas include software developers and engineers seeking hands-free coding documentation; writers, journalists, and content creators needing efficient drafting; executives and managers dictating emails and messages; and accessibility users requiring reliable voice input.
3. Use Cases: Dictating code comments and documentation directly in an IDE like VS Code; sending quick, polished messages in team chat apps without typing; drafting long-form content or emails; providing real-time captions or notes during meetings; and enabling computer use for individuals with repetitive strain injuries (RSI).

Unique Advantages

1. Differentiation: Unlike subscription-based commercial apps (e.g., Wispr Flow), Blurt is free and open-source. Compared to built-in macOS dictation, it offers significantly lower latency, higher accuracy (reporting up to 30% fewer hallucinations than models like Whisper), and universal application compatibility.
2. Key Innovation: Its tight integration with the AssemblyAI Dictation API is its core innovation. This provides a direct pipeline to a production-grade, continuously-improved speech model with real-time LLM cleanup, a combination not found in other free or open-source dictation tools. The "hotkey-hold" interaction model also optimizes for spontaneous, interrupt-driven dictation.

Frequently Asked Questions (FAQ)

1. Is Blurt really free to use? Yes, the Blurt application itself is free and open-source (MIT licensed). However, it requires your own AssemblyAI API key. You pay only for the transcription usage through AssemblyAI (approximately $0.62 per hour), and new accounts include free credits to start.
2. How does Blurt's accuracy compare for technical terms and coding? Blurt leverages AssemblyAI's Universal model, which is trained on diverse data, including technical jargon and code. For optimal accuracy with proprietary terms, you can enhance it using AssemblyAI's Custom Vocabulary feature directly via the API.
3. Can I use Blurt on Windows or Linux? No, Blurt is currently a native application built exclusively for macOS 14 (Sonoma) and later. Its design integrates deeply with macOS accessibility and input systems.
4. How do I build my own custom dictation app using Blurt's code? The Blurt GitHub repository serves as a complete reference implementation. Developers can fork the repo to modify the UI, add features, or integrate the core audio capture and AssemblyAI WebSocket streaming logic into their own projects.
5. What are the system requirements and latency I can expect? You need a Mac running macOS 14+. Latency is primarily dependent on the AssemblyAI API, which delivers first transcriptions in roughly 150ms, with full cleaned text returned in under a second. Your network connection and audio length will affect real-world performance.