Help

MurText converts audio and video files on your computer into text.

• It’s 100% free. “Free” is not used as a marketing gimmick.
• Offers unlimited transcription.
• No need to upload your files anywhere.
• Doesn’t require an internet connection for transcription.
• Focuses on results. Always prioritizes outcome.
• Simple and user-friendly design is accessible to everyone.
• Recognizes and transcribes over 100 languages.
• You decide which model to use.
• Right-click menu method is practical and efficient.
• Supports multiple languages.

To use the speaker recognition and alignment feature, you must have an access key. Access keys provided free of charge by Hugging Face must be authorized on 2 models. You can create a Hugging Face access key by following the steps below.

1. Open https://huggingface.co/login. If you have already registered, log in directly or create an account.
2. Open the "Profile" menu. Click the "Settings" link. Click the "Access Tokens" heading on the page.
Note: After clicking the link, options appear below on the same page. If you are using a screen reader, you can navigate using the level 3 heading shortcut.

3. Click the "Create new token" link.
3.1. Fill in the "Token name" field.
3.2. Check the checkboxes on the page and click the "Create token" button.
3.3. Copy the access key provided to you. You need to grant access to 2 models for the key to work.

4. Go to https://huggingface.co/pyannote/speaker-diarization-community-1.
4.1. Find the "Agree and access repository" button on the page.
4.2. Just below the button, there are two form fields for statistical purposes. Fill in the "Company/university" and "Use case" fields.
4.3. After filling in the forms, click the "Agree and access repository" button at the bottom.

5. Go to https://huggingface.co/pyannote/segmentation-3.0.
5.1. Find the "Agree and access repository" button on the page.
5.2. Just below the button, there are two form fields for statistical purposes. Fill in the forms and click the "Agree and access repository" button at the bottom.

Yes. MurText can create subtitle files with .srt and .vtt extensions. You just need to send the file to the application and make the necessary selections.

Yes. You can enable this feature using your free access key provided by Hugging Face. During the transcription process, speakers will be identified with different names. Afterwards, you can rename the speakers if you wish.

• Your privacy is 100% guaranteed.
• Your files are not used in model training.
• No internet dependency.
• No time wasted uploading files to cloud servers.
• You don’t share your transcription texts.
• No need to clean up sensitive texts or history from data collection.
• Fast results.
• Save your transcription files wherever and however you want.
• You get the performance you paid for when buying your processor! ☺

Spelling errors arise from incorrect transcription of sounds. You can use the large-v3 model to minimize spelling errors during transcription. After transcription, you can correct these errors with the help of artificial intelligence. Claude shows the best performance for this type of editing. When providing the .srt file to an AI, do not forget to give the following strict instructions:
• Never touch the timestamps.
• Treat timestamps as numerical (integer) values. Never treat them as strings.
• Do not use characters that cause problems in CMD, such as ':', '\', '%' if they appear in the text. Do not try to handle them with escape characters.
• Correct foreign-language names and technical terms found in the text.
• At the end of the process, check whether there are any time overlaps between segments.

MurText is not a commercially motivated software. We use our limited financial resources for server services and our other projects. For this reason, the software is not signed with an SSL certificate. Körüz.biz is a non-governmental organization that has been active since 2014. You can scan the software with any antivirus application you prefer. They contain no harmful content. Windows displays these warnings for all software that is not digitally signed, as this is its standard approach.

MurText generates subtitle files in two different locations during the transcription process.
Files located inside the project directory are shaped according to the settings in the software.
Files under Project directory\fallback are raw files generated directly during the transcription process and have not been processed.
If you plan to edit the generated .srt file later in an SRT editor, you can use the .srt and .vtt files under fallback\. This is because these files do not undergo processing such as setting the number of characters per line or line splitting.
If you want to spell-check the .srt file and use it directly, you can prefer the files under the project directory.

• Download the setup file from the site menu and follow the instructions.
• Choose the app language and the language of the files you’ll mostly transcribe.
• Download the model package you want.
• When needed, open MurText or right-click your file and select “MurText Speech-To-Text.”
• Transcription will begin based on your preferences.

If your goal is to generate subtitle files, we recommend using the WhisperX engine. In this engine, word-level start and end times are calculated, resulting in much more realistic timestamps. In the Fast-Whisper engine, subtitle segments are generated using the timestamps returned from the model. WhisperX performs better in time intervals with long silences, background noise, or music.

This varies depending on the transcription engine you choose and the quality of the file. With the WhisperX engine and a file of average audio quality, the success rate is in the range of 91-96% (excluding minor spelling errors and deviations of 0.2 seconds).

There are two possible reasons for this: 1. The technical specifications of your graphics card are insufficient, or you are using an older graphics card model.
2. Your graphics card is new but the AI models used by MurText have not yet been configured for these cards. (Example: The NVIDIA 5060 model can be actively used in the transcription model. However, the word alignment model does not yet support the sm_150 architecture. If an unsupported process is encountered, transcription continues with the CPU.)

No, you can transcribe as many files as you want with MurText.

No, there’s no duration limit. However, we recommend paying attention to certain points. (See: What should I consider during transcription?)

• Select the correct language of your file. Wrong selection may slow down or distort the transcription.
• If you don’t know the file’s language, use the “Auto-detect X Language” option.
• For files longer than one hour, splitting them into parts speeds up the process.
• Model selection directly affects quality and speed. If minor errors aren’t critical, choose the most efficient model.

Whisper is a speech recognition model developed by OpenAI. It recognizes many widely spoken languages and transcribes audio files in those languages.

Whisper consists of .pt format recognition packages. The audio file is sent to the selected model, which splits it into segments and begins decoding.
Since multiple segments are processed simultaneously, transcription time is shorter than the file duration. Once all segments are transcribed, they’re merged and returned.

No, your files are transcribed without leaving your device. The process uses only your computer’s processor and hardware.

No. MurText only requires an internet connection when downloading models and checking for new versions.

No. Not even error logs are sent from your device.

Transcription time varies depending on the size and quality of the file and the hardware specifications of your device. If you have an NVIDIA graphics card with CUDA support, transcription is performed faster with GPU support. If you do not have an NVIDIA graphics card or have an unsupported model, processing is done with CPU.

Yes, you can continue using your computer.
Normally, your OS manages power sharing automatically.
However, if you’re transcribing a long file with a heavy model, playing high-performance games may slow down the process.

MurText works on Microsoft Windows 7 and later.
Windows 11 was used for accessibility and performance testing.

A macOS version is planned for future releases.

.aac .avi .dat .flac .flv .m4a .m4v .mkv .mov .mp3 .mp4 .mpeg .mpg .ogg .opus .ts .wav .waptt

Make sure you are using the latest version of the application. Reinstall the latest version and uncheck the "Keep application settings" option specified during installation. If the problem persists, remove the Whisper model you are using and download it again. If the result does not change despite doing all of this, contact us using the "Report Error" button.

Go to the "Models" page in the application settings. Click the "Remove" button next to the model you want to process. Once the button switches to the "Download" position, download the model again.

• Make sure you have an active internet connection before updating.
• If you modified the config file, you may get errors. Delete its contents and restart the app.
• If you shut down your device during update, reinstalling the app is the easiest fix.

Go to Settings > Choose your app language and save.
If you want the interface in your native language, translate the “default.po” file in the app folder and send it to us.