Speech-to-text transcription with NoScribe

Introduction

The tool combines various open source technologies in a graphical user interface to create quality transcriptions.

Whisper AI developed by OpenAI is used in the background, as well as faster-whisper developed by Guillaume Klein and Pyannnote developed by Hervé Bredin for speaker identification functionality.

It is possible to identify different speakers with S00, S01, S02 etc.

Time codes are added to the transcription.

Stop words and laughter and "Euhm's" are skipped and pauses are marked with "(..)".

Some parts of a recording may therefore be lost; for some researchers these elements are important. It is therefore important to review the transcriptions and possibly correct errors, as no automatic transcription is flawless.

After transcribing, it is possible to edit the transcription in an integrated text processor with the ability to listen to the audio in the background to compare the transcription with the recording.

NoScribe can be installed on Windows, Mac or Linux. Automatic transcriptions are performed offline; no internet connection is required after installation. A potential disadvantage is that transcribing can be very slow on old or underperforming computers. The more powerful the computer, the faster the transcription will go. On a relatively powerful laptop it was possible to transcribe a 10-minute audio recording with different speakers in Dutch in 6 minutes.

NoScribe is currently being actively developed, so bugs are not excluded. For example, on Windows, the "Start" button may not be visible when opening the program; it suffices to simply enlarge the window downwards to make the part visible/usable.

Whisper AI, the automatic transcription system in the background, can tend to hallucinate during silences, interpreting the background as text.

Multilingual audio recordings are not supported.

The Speaker Identification function is not 100% accurate and may sometimes note more speakers than there actually are.

Installation

NoScribe works on Windows, MacOs and Linux.

Windows

On Windows there are two choices, depending on whether you have a computer with an integrated graphics card (GPU) or a separate Nvidia graphics card.

NoScribe can perform transcriptions faster using your Nvidia GPU; only Nvidia graphics cards are supported.

To check if your computer has a separate nvidia graphics card, open task manager with ctrl+shift+esc and navigate to the "Performance" menu.

Look here for an Nvidia GPU brand. Check if the video memory or VRAM is 6GB or higher, found? Then continue with the installation instructions from here.

Otherwise proceed with the following installation instructions.

For computers with an integrated gpu

Go to and click on the link under "The general purpose version for normal PCs without a NVIDIA graphics card"

Normally this is a "drive.switch" link to an online folder.

Click on the "noScribe_setup_0_5_1.exe" link, a download should start.

After downloading (which may take some time) you can install noScribe like any other software. Starting the installation may take some time.

If you get a message "Windows protected your PC" click on "Run anyway". Windows is often suspicious of open source software, in this case it is a false positive and noScribe is safe to use.

• There is a separate bug with this version of noScribe that you potentially can run into, luckily its easy to solve.

The start screen of noScribe gets scaled incorrecty so the screen looks like:

Instead of:

• Click and drag the bottom part of the noScribe window to make the start button visible.

(This problem has been reported on the noScribe github, the developers are aware of the problem)

Up to here with the installation of noScribe.

For computers with a dedicated GPU

Follow this link and click on the link under A special version using CUDA acceleration on NVIDIA graphics cards with at least 6 GB of VRAM:.

• Click on both files on the website "noScribe_Setup_0_5_cuda.exe" and "noScribe_setup_0_5_cuda.nsisbin" to download them.

This may take a long time, especially the nsisbin file is large.

• To successfully use the separate graphics card when transcribing, a correct installation of Nvidia CUDA toolkit must also be present. This ensures that noScribe can utilize the power of the GPU to speed up the transcription process.
• To install the correct CUDA version we need to find the GPU Driver version.
• Search for Nvidia control panel and press enter to open the program.

• Then click on "Help" and then on "System Information"

• In the following menu "System information" under "Details" there is a line "Driver version:"

• Compare now the Driver number with this table to know which CUDA version is compatible with your graphics card.
• Then go to this website to download the correct CUDA version.

When all noScribe files and the CUDA toolkit files are downloaded:

1. First install the Nvidia CUDA toolkit.
2. Restart your computer
3. Make sure the noScribe_setup_0_5_1.exe setup file and the nsisbin file are located in the same folder
4. Double click the noScribe_setup_0_5_1.exe file, it may take a while before the setup starts, be patient.

When the installation of noScribe is completed, open noScribe.

* There is a separate bug with this version of noScribe where you potentially can run into, luckily it is easy to solve.

The start screen of noScribe gets scaled incorrectly so the screen looks like:

Instead of:

• Click and drag the bottom part of the noscribe window to make the start button visible.

(This problem has been reported on the noScribe github, the developers are aware of the problem)

Up to here with the installation of noScribe.

Mac

Depending on the type of processor and the version of your operating system (MacOs) you need to download and follow a specific installation procedure.

Tp identify the processor of your computer, click the Apple icon at the top left and select "About this mac".

In the following menu next to "Chip" or "Processor" there should be a model.

For a Mac with an "Apple MX" chip:

If your computer has an Apple MX chip, start here with the installation procedure here

If your Mac is older (older than 2020), there is a good chance you will find a "Intel Processor" listing.

In case your mac is equipped with an Intel processor, also check which macOS version runs on your computer.

For intel computers with macOS Sonoma (14) or Sequoia (15) start here with the installation procedure

If you are using macOS macOS 11 (Big Sur), 12 (Monterey) or 13 (Ventura) start here with the installation procedure

For new Macs with an M1, M2, M3 or M4 processor (CPU).

• Download this installation file (noScribe_0.5.0b_arm64.dmg).
• Double-click the downloaded dmg file and drag noScribe and noScribeEdit to the link in your Applications folder (with the label drag both here to install).
• You need Apple Rosetta2 Intel Emulator ( more info) because FFmpeg, a component in noScribe is not yet compatible with the Mx CPU's. FFmpeg is used by noScribe to convert video and audio files to the .WAV format to facilitate transcription.
• Install Rosetta2 manually,
• Open Terminal with cmd + space, then search for "Terminal" and press enter. Or navigate to the /Applications/Utilities/ folder and open the "Terminal.app".
• With the terminal opened copy the following command softwareupdate --install-rosetta or softwareupdate --install-rosetta --agree-to-license and press enter to execute it.
• Then follow the instructions on screen to install the rosetta2 software from apple.
• Finally, double click the noScribe application in the Applications folder to start the noScribe program.

For older Macs with an intel CPU and macOS Sonoma (14) or Sequoia (15)

Go to this page to download the correct installer (noScribe_0.5.0b_x86_64_unsigned.dmg).

Message from the developers: "Unfortunately, we are currently unable to properly sign the x86_64 package, so you will get a warning that noScribe and noScribeEdit are from unregistered developers. You must manually allow noScribe and noScribeEdit to run if your Gatekeeper is active."

So you need to reassure your computer that the software is safe enough to open and use.

• Double-click the downloaded dmg file
• Drag noScribe and noScribeEdit to the link in your Applications folder labeled "drag both here to install", you can also open your Applications folder in a new Finder window and drag both files there.
• Now double-click noScribe in your Applications folder as if you were going to start the program.
• An error will appear stating that noScribe is from an unregistered developer
• Go to System Settings, Privacy & Security/Privacy and Security, scroll down to see a message indicating that starting noScribe was avoided and click "Open anyway".
• Do the same with noScribe Editor, Double-click and normally you should also get a notification
• Go back to the same section in System Settings, Privacy & Security/Privacy and Security, scroll down to see the message stating that noScribeEdit was not started and click "Open anyway".

From now on both programs should open just fine when you start them.

For older Macs with an intel CPU and macOS 11 (Big Sur), 12 (Monterey) and 13 (Ventura)

Open this link and download the installation file (noScribe_0.5.0_x86_64_unsigned_legacy.dmg).

Message from the developers: "Unfortunately, we are currently unable to properly sign the x86_64 package, so you will get a warning that noScribe and noScribeEdit are from unregistered developers. You must manually allow noScribe and noScribeEdit to run if your Gatekeeper is active."

So you need to reassure your computer that the software is safe enough to open and use.

• Double-click the downloaded dmg file
• Drag noScribe and noScribeEdit to the link in your Applications folder labeled "drag both here to install", you can also open your Applications folder in a new Finder window and drag both files there.
• Now double-click noScribe in your Applications folder as if you were going to start the program.
• An error will appear stating that noScribe is from an unregistered developer
• Go to System Settings, Privacy & Security/Privacy and Security, scroll down to see a message indicating that starting noScribe was avoided and click "Open anyway".
• Do the same with noScribe Editor, Double-click and normally you should also get a notification
• Go back to the same section in System Settings, Privacy & Security/Privacy and Security, scroll down to see the message stating that noScribeEdit was not started and click "Open anyway".

From now on both programs should open just fine when you start them.

Linux

There are no official installation instructions for Linux yet, for now these are the instructions for Ubuntu: set up by someone who has gotten noScribe running on Ubuntu.

Note: The procedure below uses a Python virtual environment.

Make sure that Python and pip are installed.

1. Extract the latest NoScribe source code from https://github.com/kaixxx/noScribe/releases.
2. Extract the latest NoScribeEdit source code from https://github.com/kaixxx/noScribeEditor.
3. Move the noScribeEdit folder into the noScribe folder you downloaded earlier and rename the noScribeEdit folder to noScribeEdit.
4. Open a terminal and navigate to the noScribe folder (cd in the noScribe folder).
5. Run the following commands:

1. Download all files from https://huggingface.co/guillaumekln/faster-whisper-large-v2/tree/main and place them in the models/faster-whisper-large-v2 folder within your noScribe folder.
2. Download all files from https://huggingface.co/guillaumekln/faster-whisper-small/tree/main and place them in the models/faster-whisper-small folder within your noScribe folder.
3. Edit the file noScribe.py as follows to be able to open NoScribeEdit from NoScribe:

*Edit line 566 to read as follows:

*Edit line 578 to read as follows:

*Edit line 580 to read as follows:

Start NoScribe:

1. Open a terminal and navigate to the directory where the noScribe source files are located.
2. Activate your virtual environment with:

1. Run the following command:

If you have a HiDPI display, the NoScribe window and text may be very small.

You can add the following lines to the file noScribe.py after line 70:

Usage

Open noScribe, this can take some time before you see an interface, be patient, if noScribe starts .|check if you can see the blueStart button, if not, enlarge the window a bit downwards.

1. "Audio Bestand:" click the blue folder to select your audio or video file that needs to be transcribed, video files will automatically be converted in the background to .wav (without overwriting the original file).
2. "Save Transcript as:" click the blue folder to select the destination folder for your transcript and select the format, html to edit your transcript in the built-in editor, .txt to save your transcript as a standard text file or .vtt to create a subtitle file.

   

1. Select .html if you want to edit the transcript in the built-in editor.
3. "Start (hh:mm:ss):" indicate from which time in the source file the transcription should start, hh= hour, mm= minutes, ss= seconds.
4. "Stop (hh:mm:ss):" indicate until what point in the source file the transcription should be carried out, making a transcript of only the first 5 minutes can for example be done with: "00:05:00" in this field.
5. "Language:" select the spoken language of the source file, use the "Auto" option to let Whisper analyze which language is spoken. 100 languages are supported by Whisper. Not every language will yield an equally accurate transcription.
6. "Quality:" indicate how accurate the transcript should be, "Precise" takes longer but is more accurate, "Fast" is faster but the transcript will contain more errors.

   

7. "Mark pause:" mark pauses longer than x number of seconds. The options are, "None", do not mark any pauses, 1sec+, longer than 1 second, 2sec+, longer than 2 seconds, 3sec+, longer than 3 seconds. Pauses are transcribed as parentheses with one dot per second in between, for example '(..)' for a pause of two seconds. Pauses longer than 10 seconds are written out as '(XX second pause)' or '(XX minute pause)'. You have the option to mark pauses of one second and more ('1sec+'), two seconds and more ('2sec+') or only the longer pauses of three seconds and more ('3sec+'). Choose 'none' to disable this feature completely.

   

8. "Speaker detection:" options to predefine some things about the number of speakers in the source file, indicating the number of speakers beforehand can make it easier for Pyannnote to identify the speakers. The "Auto" option will try to identify the number of speakers, no option here is foolproof. It is possible to set the number of speakers to "None" if the function of identifying speakers is not needed, which halves the transcription time; the transcript becomes a continuous block of text.

   

9. "Overlapping Speech:" indicate whether there are speakers speaking over each other in the source file or not, if this option is on, noScribe will try to mark moments when two people are talking at the same time with //double slashes//. (this is an experimental feature.)
10. "Timestamps:" when enabled, noScribe inserts timestamps in the format [hh:mm:ss] into the transcript, either at each speaker change or every 60 seconds.
11. "Start:" The Start button, or "Cancel" button when a transcription is being made, this button tends not to be visible when starting noScribe.
12. "Editor:" The button to open the built-in text processor and, if necessary, adjust the transcript based on the audio.
13. "Log window:" A window that gives you information about the version of noScribe, the link to the Github page. Here will also be visible the progress of the transcription.

Starting Transcription

• Navigate to an audio or video file you want to transcribe. Via the open "Audio Bestand:" option.

• After selecting the source file, select the format of the transcript and press "Save" or "Opslaan".

Select .html if you want to edit the transcript in the built-in editor.

• Indicate via the Start and/or Stop option which part of the source file you want to transcribe, leave the options as they are if you want to transcribe the entire file.
• Select the language, if you know it.
• Choose the quality of the transcription using "Quality:"
• Decide whether pauses longer than x seconds should be noted
• If you know how many speakers there are in the source file, indicate this number here.
• If there are overlapping speakers in the source file or not, indicate this in the "Overlapping Speech:" option.
• Indicate whether you want to see timestamps in the transcript.
• Then click "Start".
• The transcription will now start, which may take some time depending on how powerful your computer is, how long the audio/video file is and how many speakers there are, how precise the transcription should be. The progress should be visible in the log window with a red progress bar at the bottom

Adjusting Transcription

• When the automatic transcription process is complete, you will see the following screen.

• Normally the editor will open automatically if that doesn't happen click on the gray "Editor" button in the bottom right of the log window.

• In the editor you can adjust the transcript. It is possible to click somewhere in the text and play the relevant piece of audio by clicking on the "Play/Pause Audio" button.

• The text will be highlighted during playback of the audio.
• Save the edited text after editing

Editing an earlier made transcription in the editor

This only works if you have saved the previous transcript as html.

• Open the noScribe editor, this is a separate piece of software. On Mac it's in the Application folder
• On windows search for "noScribe Editor" via the start menu.

• In the editor use the map icon at the top left and select a previously made transcript.

• It may take some time for a transcript to load, depending on its size

• If you don't have the original source file in the original location, the function of playing the audio simultaneously with the transcription will not work. If the source file has not been moved, audio will play in the background when you click the "Play/Pause Audio" button.

That concludes the manual for noScribe, it is important to note again that the software is under full development and the manual was written at an early stage of that process. The information in the manual may therefore be outdated, check the sources via the official Github of noscribe, you can always create an account on Projectcest to contribute to manuals like this.

If you have problems with noScribe please let us know at the issues page of the Github page of noScribe. Mention which computer you are using, with which version of your operating system and there is a good chance that the developers can solve your problem or answer your questions.