Introduction

Welcome to the official website of DuRT.

The name DuRT means Do Speech Recognition and Translation in Real-Time, enabling real-time speech recognition and translation.

Use Cases

DuRT can recognize both internal and external audio on the device, converting speech into text and translating it into a specified language. Internal and external audio refers to audio played on the device as well as audio received via the microphone.

When watching online videos, DuRT helps you overcome language barriers by recognizing the speech in the video and translating it into your native language in real time.

During online classes or meetings, DuRT helps you understand the speaker in real time, with the option to revisit the recorded audio or text transcription afterward.

When watching local videos, you don’t need to search for bilingual subtitles. DuRT can translate the speech in the video in real time, helping you understand its content.

DuRT also supports saving the recording, recognition results, and translation results.

Installation

Install from the Mac App Store

Key Features

Speech Recognition

DuRT can recognize both internal and external audio on the device in real time, converting speech to text. It displays the recognized text in a floating window, as shown below:

Currently, DuRT offers three types of speech recognition: Streamed Speech Recognition, Non-streamed Speech Recognition, and Apple Recognition.

• Streamed Speech Recognition converts speech to text in real time.
• Non-streamed Speech Recognition converts speech to text with a few seconds’ delay (configurable), closely simulating real-time recognition.
• Apple Recognition utilizes macOS’s built-in speech recognition to convert speech to text in real time.

Streamed Speech Recognition supports four languages: English, Chinese, Korean, and French.

Non-streamed Speech Recognition supports the following languages: Arabic, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Finnish, French, Galician, German, Greek, Italian, Japanese, Korean, Macedonian, Polish, Portuguese, Romanian, Russian, Slovak, Spanish, Swedish, Tamil, Thai, Turkish, Ukrainian, Urdu, and Vietnamese.

Apple Recognition supports Arabic, Cantonese, Catalan, Croatian, Czech, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Malay, Mandarin, Polish, Portuguese, Romanian, Russian, Shanghai dialect, Slovak, Spanish, Swedish, Thai, Turkish, Ukrainian, and Vietnamese.

Based on a comparison of these three methods, you can choose the one that best suits your needs.

Text Translation

DuRT supports text translation by using a local AI model to translate recognized speech into the specified language. It displays the translation in a floating window, as shown below:

Supported languages: Arabic, Bulgarian, Catalan, Chinese, Croatian, Czech, Danish, Dutch, English, Finnish, French, Galician, German, Greek, Italian, Japanese, Korean, Macedonian, Polish, Portuguese, Romanian, Russian, Slovak, Spanish, Swedish, Tamil, Thai, Turkish, Ukrainian, Urdu, and Vietnamese.

Save Audio and Recognized Text

Before starting recognition, you can select the corresponding options to save the audio, recognized text, and translated text for the entire recognition process.

Copy Recognized Text

Click the text in the floating window to enter a red highlight mode. The text is copied to the clipboard automatically, allowing you to paste it elsewhere using command+v. As shown below:

Hardware and System Requirements

Currently, DuRT is available only for Mac computers. It requires macOS 13.0+ and supports only Apple silicon (M-series chips).

Streamed and non-streamed recognition require downloading an AI model and use memory while running, approximately 2-3 times the model's size.

Apple Recognition uses macOS's built-in functionality, so it doesn't require additional memory.

Translation requires downloading an AI translation model, with an approximate memory requirement of 1.5GB.

Most Mac M-series computers have at least 8GB of memory, sufficient for this requirement.