WhisperScript, an Electron desktop app GUI for Whisper

[jonathgh]

Thanks to the work of @ggerganov and with inspiration from @jordibruin, @kai-shimada and I were able to implement Whisper in a desktop app built with the Electron framework. The app runs on Mac at the moment, but we hope that Electron will also allow for cross-platform compatibility in the future. You can download it here: Whisperscript

Currently, our features include:

• Audio playback and navigation from the transcript
• Transcript-level search
• Segment-level editing
• Storing and re-opening transcripts from SRT files
• Batch imports
• MP3, WAV and M4A formats
• Exporting to TXT, CSV, SRT and VTT
• Both Intel and Apple Silicon Macs

We hope you enjoy it and let us know if there are any other features you would like to see. You can reach us at [email protected]

[tonyho]

@jonathgh Are you going to make it cross-platform for Linux + Windows?
Or is there a github repository for this project?

[jsmollin]

Why are you choosing not to create a Linux version?

[candideu]

If you're looking to use Whisper this way on Linux: Kdenlive's (video editor) latest release integrates Whisper to create subtitle files. It's FOSS

[jsmollin]

If you're looking to use Whisper this way on Linux: Kdenlive's (video editor)...

Thank you for this suggestion. My interest in this particular application is the aggregate of features the author has put into the same application.

[qhkm]

I've made a similar gui apps compatible for both windows and mac. Will share once it's done

[jonathgh]

@qhkm We'd love to see what you have made once you are done with the app! Send us a mail, maybe we can send some people your way who are looking for the Windows or Linux versions

[kai-shimada]

👉 UPDATE: WhisperScript now added support for MKV, MP4 and MOV Video Import!

[bbecausereasonss]

Is it possible to utilize Whisper Jax in this GUI?

https://github.com/sanchit-gandhi/whisper-jax

[kai-shimada]

interesting, do you know how it performs compared to whisper.cpp?

[bbecausereasonss]

Up to 70x faster.

[August78]

Wow this is great! Subbed the topic, and waiting for a windows client :)

[jonathgh]

We hear you! It is on the roadmap :)

[nhan000]

You can use a desktop GUI app on Windows here https://github.com/Const-me/Whisper

[Makememo]

Maybe try https://memo.ac/, which supports Windows GPU.

[candideu]

@Makememo How can users obtain an invitation code?

And when will the docs be available in English?

[Makememo]

@candideu sorry, These are some codes.

WEAE-HPeM-uhR8-nuk6
E9sh-536K-38n7-LdbL
95Ac-r667-WSFZ-nqis
M9UY-ZuHA-4zej-fxb8
Ck7C-vmEf-Ugeg-pifn
j43j-enHE-oqPo-bBfB
N7GF-VVRU-Pcja-5Cks
mape-7ssz-Xj2A-P5Wf
5BRJ-fxAf-pfvA-ZcVZ
LAL9-pjfR-aD9M-7BXf

[mrgalindo]

It would be great if you could add speaker recognition! i would get it right away!

[jonathgh]

Speaker recognition would be great to have – we are looking into it, as it's quite a requested feature. We would also love to include speaker labels in the transcripts. I know that there have been several proposals to use Pyannote to diarize the segments.

[zenminimalist]

Yupp, same here. Would LOVE that!

[duttaoindril]

Why is the paid feature for better whisper models? It should be paid for features you've written yourself.

[jonathgh]

We totally agree - we are about to release our latest version, which includes a lot of editing, visualization, favoriting and exporting features which are features built on top of the models. We think that these features are worth the premium price.

[jonathgh]

@duttaoindril how do you like the new update? Let us know how you're using it, we'd be happy to hear from you!

[duttaoindril]

That's great, and I really appreciate those features.

But the larger models are still behind payment gates, it'd be better to strip down the free experience but offer all models in the free version.

A free trial of a few days or # of transcriptions with the pro features would also help you make sales. Along with an auto updater with constant feature drip for pro users.

Pro feature ideas:

• try to find ways to improve the performance of the models, and sell that in the free version - don't artificially slow down performance
• when dragging in a video, also show the video as you scrub through the transcript and audio waveform
• after adding video add speaker diarization to try to automatically label voices and faces if any in the UI
• after adding diarization make it easy to export all transcriptions as time and person labelled subtitles to add them back to sites like YouTube or movies or anything.

Point is, there're a bunch of features you can work toward putting together to make this an increasingly valuable and enticing paid product - I just don't think you should payment gate the larger models since it's open source and not your work.

[jonathgh]

@duttaoindril Glad you are enjoying the new features. Thank you for the feature ideas, we really appreciate them and are taking user feedback seriously when prioritizing new features. An auto-updater and feature drip for the Pro users are both on the roadmap. Diarization is something we are also working on, but will take a few releases to get right.

As for the paywall, I think there is a bit of a misunderstanding. We aren't paywalling any models, the models are totally free to use and open for everyone to download. We are simply only offering the GUI and transcript editing / visualizations features for the larger models to Pro users, which we did write, and will continue to improve.

[jonathgh]

We're excited to announce WhisperScript v1.2.1, an update to our Electron desktop Whisper implementation that introduces a lot of new features to speed up your transcription workflow. This update adds a bunch of improvements to the visualization, playback, editing, and exporting of your transcripts. Here's what's new in v1.2.1:

• Multiple selection and segment actions (copy, remove, merge, bookmark)
• Bookmarking, filtering, and exporting bookmarked or selected segments as individual clips.
• Enhanced transcript functionality (auto-saving changes, copy without timestamps)
• Improved audio format support (FLAC, OGG, OPUS)
• Introducing Regions, transcript segment visualizations on the timeline – to see exactly where the text is being spoken.
• Add new Regions, Adjust Region length, move Regions around, merge and delete Regions.
• Optimizations for better usability and performance

We hope that these new features will speed up your workflow and make it easier to edit and navigate your transcripts. We’re actively developing more, and we're excited to see how you'll utilize these features in your projects!

Download the latest version here. The features above are only in the Pro version: https://getwavery.com

We'd love to hear your thoughts and suggestions for future updates. You can reach us at [email protected].

Happy transcribing!

[sscotti]

Might have to try it. BTW, I started playing around with Whisper in Docker on an Intel Mac, M1 Mac and maybe eventually a Dell R710 server (24 cores, but no GPU). Not sure you can help, but wondering about mutli-CPU and/or GPU support in Whisper with that hardware. It sounds like it might be partially possible, but NVIDIA GPU's are the only ones that are supported much.

I want to integrate the thing into a medical IT application stack that I have, just using the Whisper API in the local build. I have an OpenAI API Key for testing also.

[jwnacnud]

Totally agree. Find ways to speed up the larger models. That’s worth paying for. Hey people hooked on the larger models, even though slow, and also offer the smaller and faster models. Your service and application are your selling points, not the language models.

On Sun, Mar 26, 2023 at 1:13 PM Oindril Dutta ***@***.***> wrote: That's great, and I really appreciate those features. But the larger models are still behind payment gates, it'd be better to strip down the free experience but offer all models in the free version. A free trial of a few days or # of transcriptions with the pro features would also help you make sales. Along with an auto updater with constant feature drip for pro users. Pro feature ideas: - try to find ways to improve the performance of the models, and sell that in the free version - don't artificially slow down performance - when dragging in a video, also show the video as you scrub through the transcript and audio waveform - after adding video add speaker diarization to try to automatically label voices and faces if any in the UI - after adding diarization make it easy to export all transcriptions as time and person labelled subtitles to add them back to sites like YouTube or movies or anything. Point is, there're a bunch of features you can work toward putting together to make this an increasingly valuable and enticing paid product - I just don't think you should payment gate the larger models since it's open source and not your work. — Reply to this email directly, view it on GitHub <#1028 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGW5A736VDBX25WQ4VFS43W6CILZANCNFSM6AAAAAAVPS7CDI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- Jeffrey Duncan

[leidmew]

Hi! Just update to PRO version because I think there's value added for people like me without coding experience.. Just one feature I'm missing and that may be easy to implement (speaking with zero idea so enlighten me if otherwise) is to be able to decide where the models are downloaded. By default a new folder is created in Documents folder (I'm on Mac) so it's annoying to have a high level folder in there just for that. Thanks for your efforts with the app!!

[jonathgh]

Thanks for your support! We have had some other requests for this feature, and we will try to get this into a future update. I know it can be a big download to keep it on the main drive. Thanks for the feedback.

[bbecausereasonss]

Can this be updated to use https://github.com/guillaumekln/faster-whisper

[jwnacnud]

I saw WAY faster speeds but also more hallucinations. I'm ok with slower, if it means more accurate. Jeffrey Duncan

…

On Mon, Apr 24, 2023 at 12:07 PM becausereasons ***@***.***> wrote: Up to 70x faster. — Reply to this email directly, view it on GitHub <#1028 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGW5A5I2MGI4AV5RHMZPC3XC253BANCNFSM6AAAAAAVPS7CDI> . You are receiving this because you commented.Message ID: ***@***.***>

[kai-shimada]

I agree. I think we'll stick with whisper.cpp. It has also received performance improvements through added support for CoreML (for Apple silicon devices) and 4-/5-bit integer quantization in their latest v1.4.0 beta release. We'll definitely try to implement these improvements once a stable version is available.

[Veneration1]

Is model large-v2 implemented or going to be implemented? Here's what that page says about it: "The large-v2 model on average shows about 5% relative error reduction in English and about 10% in other languages, but please note that it may behave differently depending on the individual audio and in some cases perform worse than large-v1." Here it is on GitHub it's also on Hugging Face

[kai-shimada]

Hi, the large model we use is the large-v2. In the upcoming release, we'll also start supporting the english-only models.

[Veneration1]

A few questions:

• Would be nice to be able to have the option to export a txt without timestamps. I see you can copy the transcript and paste it to do that, but would be nice to have a simple toggle for that and the normal export.
• Without knowing if the large model will run on my computer I'm hesitant, as it may not run or may be too slow for our needs, like 3+ hours rather than 1-2 hours to work 1 hour of audio. We need a large model quality for our work. So wondering if there are any simple ways to test this. Does it use GPU? My Mac has intel.
• I see you can modify text in-app, which is cool. It would probably be very difficult to code, but I've been looking for an app or way for content we edit to be learned so that we don't have to keep making the same corrections recording after recording. Could almost be a different app, maybe there is one..?
  Thank you for your reply to my other question. Wishing you all the best on your continued development. SUPER excited to have speaker identification in the app. Very much looking forward to that.

[kai-shimada]

• Would be nice to be able to have the option to export a txt without timestamps. I see you can copy the transcript and paste it to do that, but would be nice to have a simple toggle for that and the normal export.

I'm currently working on the new export window which will allow you to do that.

• Without knowing if the large model will run on my computer I'm hesitant, as it may not run or may be too slow for our needs, like 3+ hours rather than 1-2 hours to work 1 hour of audio. We need a large model quality for our work. So wondering if there are any simple ways to test this. Does it use GPU? My Mac has intel.

I have an old MacBook Pro from Mid 2012 and it takes about 2 times the audio duration to transcribe using the large model (2h~ to transcribe 1h of audio). If you are using a newer Intel Mac, it should transcribe much faster. I assume you don't have the pro license, in this case, the simplest way to test it out would be to go to https://github.com/ggerganov/whisper.cpp#quick-start, follow the quick start guide and let it run from your terminal. WhisperScript doesn't support GPU yet, but it seems like whisper.cpp is on its way to release a stable version soon with GPU support, so when it's out, WhisperScript will support GPU inference too.

• I see you can modify text in-app, which is cool. It would probably be very difficult to code, but I've been looking for an app or way for content we edit to be learned so that we don't have to keep making the same corrections recording after recording. Could almost be a different app, maybe there is one..?

Do you mean something like MacWhispers "Global Find & Replace" which will let you build your own "dictionary" of commonly misspelled words and their correct spellings so that these words get automatically corrected each time? We haven't implemented it yet but it will be in near future update for sure. In the upcoming release, we instead added an option to add a prompt that can be used to introduce new words.

• Thank you for your reply to my other question. Wishing you all the best on your continued development. SUPER excited to have speaker identification in the app. Very much looking forward to that.

Speaker Diarization is a very requested feature but it also needs a bit more time to implement, we'll let you know soon enough when we are ready ;)

[darnn]

@Veneration1 If you can get Subtitle Edit to work on a Mac, with Wine or something, it has a file where you can add automatic corrections:
https://github.com/SubtitleEdit/subtitleedit/
(And the file is eng_OCRFixReplaceList.xml, or whichever language you need. The format is pretty intuitive, just add your own lines.)

[shruru]

Looks interesting . Glad to see a project got updated continually 👍🏻

May I ask what's the main differences between this and MacWhisper if you don't mind? 😁
Speed / Accuracy (by same model) ?

BTW when it's working on Windows, I do hope it could supports GPU for speed-up processing.

Thank you.

[matsumurae]

At least for me, WhisperScript works faster I don't know why. Same conditions, same file, same model. Small model took 17 min MW vs 4 on WS. After seeing how slow was MW I didn't wanted to wait for the Medium, on WS took 11 min.

Note: I'm using an Intel mac, not M1. 2020 i5 if you're curious.

[darnn]

@shruru As mentioned previously in this thread, for a Windows implementation that utilizes the GPU, you might want to look into this: https://github.com/Const-me/Whisper/releases/

[edtechdev]

Another Windows app is whispercppGUI https://github.com/Topping1/whispercppGUI
I combine it with yt-dlg to generate more accurate YouTube transcripts https://yt-dlg.github.io/yt-dlg/

[jonathgh]

EDIT: We have now released a new version of WhisperScript, with several improvements, including a video player, batch processing and improved performance overall. We'd love to hear your thoughts on the new design! Join the discord to hear about our latest developments.

Exciting news! We're back with an impactful update to improve your transcription experience, making it more efficient for language learners, subtitle creators, interview analysts, or those combing through their media libraries. We're introducing WhisperScript v1.3.4, and here's what's new:

New in WhisperScript v1.3.4:

• Transcription Settings Window 🪟: Now you can view and modify imported files, limit the number of characters per segment, and translate any language to English right within the window! Currently we are supporting only any-to-English translation.
• Limit number of characters per Segment: customize subtitles by limiting the length of produced segments.
• Initial Prompting 🪄: control the output of WhisperScript to add domain-specific vocabulary, example formatting and selecting language variants (e.g. Simplified vs. Traditional Chinese).
• Auto-Updater 🌟: Always stay up to date with the latest features.
• Find & Replace All 🔍: Easily replace certain words, names, or abbreviations in the transcript.
• Model download manager: Save space and time - download multiple models in parallel, and set a custom folder to store the models (e.g. on an external drive).
• Merge segments: Combine related segments by sentence.
• AAC Audio Format Support: Support for more audio formats.
• Paste Command : Transcribe files from the clipboard right onto the WhisperScript window.
• New Transcript Export Options: Quickly output the current transcript, current selection, audio clips, document files, SRT or Anki APKG (for language learners) or current bookmarks in the new export window!
• Beta Channel Updates: subscribe to the Beta channel within the app to get the cutting-edge features.

These features are available to Pro Users, but Lite Users will also get several UI improvements and bug fixes:

• English Only Models: for faster, more accurate transcriptions if you are only working in English.
• Status Bar: displays information about the model and language used to transcribe each transcript, access settings and re-transcribe button.
• Re-transcribe button: to re-do transcription without having to leave the app.

We’re still actively developing additional features, and we’d love to invite you to join the Discord to be a part of the process, suggest features and report bugs: https://discord.gg/b9TYCgC6

Download the latest version here: https://getwavery.com

We'd love to hear your thoughts and suggestions for future updates. You can reach us at [email protected].

Happy transcribing!

[pietrosperoni]

Thanks, I love your tool.
How do we get the pro version?
Also the Discord invite is expired :-(.

Cheers.
Pietro

[kai-shimada]

@pietrosperoni Hi, the information above is slightly outdated. We are currently not offering the pro version. Please contact us at [email protected], and we might be able to help you further.

[pietrosperoni]

Hello @kai-shimada , thanks for your answer. I did send an email yesterday, just after writing this message when I realised the email was maybe the preferred route. I also added some suggestions I had. I will check if I get an answer in the next days.

Best Regards,
Pietro

[ghost]

Greate job!

[jonathgh]

Thanks!!

[JackLawrence6]

GJ,can it be used for ytb or twitch live streaming?

[bits-by-brandon]

I made a open-source alternative with basic features here. Feel free to fork, contribute, or do whatever you need!
https://github.com/bits-by-brandon/whisper-ui

[candideu]

Nice! Would it be possible to add build instructions? Not sure how to use it on Windows...

[ntovarsolorzano]

Here are the instructions. Feel free to share it, and/or copy-paste it for others to know.

$ git clone https://github.com/whisper-ui/whisper-ui.git
$ cd whisper-ui
$ yarn install
$ sudo apt-get install ffmpeg -y # For Ubuntu/Debian-based systems
$ yarn build # If there's a build script
$ yarn vite # Since it uses Vite

It worked for me.
Ubuntu 20.04 distro.

Update: When clicking the buttons nothing happens. 😢

God bless.

[SituDaMan]

Hi there, does it only transcribe, or it can do "translate" as well?

[jonathgh]

@SituDaMan It does support translation as well, but at the moment, just all languages to English.

[thebigboss9018]

When windows? T.T

[thebigboss9018]

Im interested in the whisperscript gui, but its only for mac.

[Makememo]

Welcome to try MemoAI (https://memo.ac )，it's support for Windows.

[thebigboss9018]

Actually this looks good for my usecase, I will test it later. Thanks for rec

[edtechdev]

When windows? T.T

See https://github.com/Const-me/Whisper

[jswhisperer]

https://dev.to/jswhisperer/cross-compile-a-distributed-electron-app-3nfk

[jonathgh]

whisperscript-05-search-replace.mov

Exciting update! WhisperScript v2.0 is here, re-written from the ground up with React and Typescript, with a whole new architecture and improved features. Check out what’s new in this release!

Wavery Accounts:

• Multi-Device Support 📲: Now you can install WhisperScript on up to 2 devices simultaneously, ensuring seamless access across multiple machines.

• Simplified Account Management 🔑: All licenses are now managed directly through Wavery, making account setup and management easier than ever.

• Super Fast and Responsive ⚡: We’ve completely rebuilt WhisperScript for speed and stability. With a fresh visual update, the app is now more responsive and intuitive to use.

• Video Viewer Support 🎥: Now transcribe directly from video files in formats like MOV, MP4, and MKV, and view the files in the new video player, streamlining the workflow for video and subtitle creators.

• Enhanced Toolbar 🛠️: Improved access to essential editing actions, making it easier to navigate and refine your transcripts.

• Search & Replace (Beta) 🔍: A new feature for quick text adjustments, although still in beta, making changes across the entire transcript faster.

• Merge Segments by Sentence: Combine segments seamlessly by sentence, available in Edit > Merge Segments by Sentence, for a cleaner transcript flow.

• Bookmarks 📌: Mark key points in your transcripts for easier reference and navigation.

• Real-Time Transcription Progress: Monitor the progress of your transcription as it happens, enhancing visibility and control.

• Abort Running Transcriptions: Easily stop a transcription in progress via the progress bar or by closing the active tab.

• Future Transcriptions Control: Delete tabs to individually abort or remove queued transcriptions without disrupting active processes.

• Filename Dropdown Actions 📝: New dropdown in the header to reveal the file in Finder, re-transcribe, or close the transcript. (Feedback on visibility is welcome—let us know if this action menu is too hidden!)

• Large v3 Turbo Model Support 🚀: Experience the speed of Large V3 Turbo—twice as fast as the original Large V3 model and comparable to Distill Whisper speeds.

• Expanded Export Options 📄: Export transcripts to HTML, JSON, PDF, Word, and RTF, enabling more flexibility in sharing and documentation.

• Open Recent Projects: Quickly access recent projects without navigating through file directories.

What This Means for You:

• Create Your Wavery Account: To continue using WhisperScript, simply create a Wavery account. It’s quick and easy, and unlocks all the new features.

• Pro Trial and Lite Version Changes: We’re streamlining our offerings by retiring the Lite version. Now, all users can try the Pro version for free with a trial. For former Gumroad Pro license holders, you will always be able to keep using WhsiperScript 1, and also get access to WhisperScript 2 by migrating your license code during signup or on the account page .

• New Pricing Structure: WhisperScript has been optimized for performance, enabling faster updates and additional features. Existing Pro users maintain lifetime access, while new users benefit from an improved value at our updated pricing. We’re offering a complimentary year of WhisperScript Pro to all Lite users to ease the transition.

Thank you for being part of the WhisperScript community! We’re thrilled to continue evolving the app with these improvements and more to come.

For any questions or assistance, reach out at [email protected]. We’re here to help!

---

WhisperScript v2.0 is designed to optimize your transcription workflow, whether you're working on interviews, media analysis, or multilingual transcription. As always, we appreciate your feedback and suggestions to help us keep improving!

Download the latest version here: https://getwavery.com

Have ideas or feedback? Reach out to us at [email protected] or join our community on Discord to connect with other WhisperScript users: https://discord.gg/b9TYCgC6.

Happy transcribing!