Hacker News Reader

Comment by properbrew | Hacker News Reader

https://play.google.com/store/apps/details?id=com.blazingban...

Completely free, no ads, no in-app purchases and no accounts / network required offline voice transcription.

I have also built the macOS/Windows/Linux versions which I'll also make free to download and available on my site soon (https://blazingbanana.com/).

iOS version is built and works (extremely well), just waiting for the Apple Developer signup process to complete.

Big shout out to https://github.com/mybigday/whisper.rn and https://huggingface.co/ggerganov/whisper.cpp/tree/main for making this even possible.

Any suggestions are welcome.

View on HN

On the subject of whisper being great... A few weeks ago a co-worker commented about the difficulty he'd had editing a work demo, I pointed at various jump-cutting tools that had automated what he did in the past (editing out silences). But I'd also wanted to play with whisper for a while...

So a couple of hours later I'd written a script that does transcription based editing: on the first pass it grabs a timestamped transcript and a plain text transcript for editing; you edit the words into any order you like and a second pass reassembles the video (it's just a couple of hundred lines of python wrapping whisper and ffmpeg). It also speeds up 4x any silences detected that sit within retained sequences in the video.

Matching up transcripts turns out to be not that hard; I normalise the text, split it, and then compare to the sequence of normalised words from the timestamped transcript. I find the longest common sequence, keep that, then recurse on the before/after sections (there's a little more detail, but not much). I also sent the transcription to ffmpeg to burn in as captions, because sometimes it makes the audio choppy and the captions make it easier to follow.

I know, tools have been doing this for years now. I just didn't have one to hand, and now I do, and I couldn't have done this without whisper.

That is absolutely awesome and I love hearing about the tools that people build themselves!

Honestly, the capabilities of whisper is insane, the fact that it's free and open source is really a gift. Some of the things it can do feels almost sci-fi.

If you ever decide to release it publicly please let me know, sounds like a very useful tool.

"release" is maybe too strong a word, it's not a lot of code and I don't plan to put any more effort into the nonexistent interface since it was just built for personal use. But the code:

https://gist.github.com/bazzargh/e1d2e2718af575a03206114a291...

This is very kind of you, thanks.

I spent a few hours editing a video in Davicni resolve to do this by hand. Then i found out this is a built in feature.

Couldn't find it on the Play store by searching for the name and the developer's name: if it is not just me then your app is very hard to discover.

So I am installing it through the link you provided, which directed me to a "install success" page saying "your purchase is successful" even if your app is free. Another obstacle to adoption :-)

Last, I was not informed on the page of the app' size. Seeing what it does and the time it takes to download I am afraid it could be huge? Third obstacle :-)

For me, searching for "whistle" on play store, I get the app as the third result (ignoring sponsored crap). Searching for "blazingbanana" gets me the app as the first result".

App info shows 218MB size, which I suppose is about what I'd expect for a model+app code :shrug:

Good to know, it's hard to know what real users would see in the play store and not Google just showing you what you want. Thank you for checking it out

Thank you for the feedback, I really do appreciate you taking the time to check it out and write out the comment! I'll look at adding a note about total app size in the description, it won't hurt.

As for discoverability / the "your purchase is successful" message, I'm not sure what else I can do, I've set it to free, no ads etc in Google Play. Maybe I need to hit a few more keywords for transcription so it surfaces it more.

The iOS Appstore also treats/words app installs as ‘Purchases’. Always confused my…

Pretty cool. I've downloaded and lightly tested. Works great.

I love the "free forever, no ads part..." But it obscures what the app is for. Maybe start with the "Speech to text transcription" to make it clearer.

Either way, that's just semantics. Great job

Thank you, really appreciate the kind words. I'll take a look at giving the description a bit of a once over for the next release coming soon.

It'd be nice to keep the voice recording too, as I noticed at least one thing that it transcribed wrong.

This way one can listen to the recording again, and correct such issues.

Great idea and an option I'm looking at implementing soon with the ability to reprocess with a different model if needed. Cheers for taking a look.

By the way, how does this handle conversations between two or more people?

Currently, terribly. It is on the roadmap after I've released it on all platforms.

That's very cool, I've been looking for a fully offline transcription app for quite a while. Thanks for building this! And thanks so much for providing an "import audio file" function, not just "record from mic" -- transcribing voice notes from various messenger apps is my main use case here.

Do you have an idea about supporting languages other than English?

Thank you, glad you like it!

The average model and upwards should support all languages from the whisper models by default.

I haven't tested them all so I'm unsure of the quality, however it should in theory support the following:

---

Albanian

Amharic

Arabic

Armenian

Assamese

Azerbaijani

Bashkir

Basque

Belarusian

Bengali

Bosnian

Breton

Bulgarian

Cantonese

Catalan

Chinese

Croatian

Czech

Danish

Dutch

English

Estonian

Faroese

Finnish

French

Galician

Georgian

German

Greek

Gujarati

Haitian creole

Hausa

Hawaiian

Hebrew

Hindi

Hungarian

Icelandic

Indonesian

Italian

Japanese

Javanese

Kannada

Kazakh

Khmer

Korean

Lao

Latin

Latvian

Lingala

Lithuanian

Luxembourgish

Macedonian

Malagasy

Malay

Malayalam

Maltese

Maori

Marathi

Mongolian

Myanmar

Nepali

Norwegian

Nynorsk

Occitan

Pashto

Persian

Polish

Portuguese

Punjabi

Romanian

Russian

Sanskrit

Serbian

Shona

Sindhi

Sinhala

Slovak

Slovenian

Somali

Spanish

Sundanese

Swahili

Swedish

Tagalog

Tajik

Tamil

Tatar

Telugu

Thai

Tibetan

Turkish

Turkmen

Ukrainian

Urdu

Uzbek

Vietnamese

Welsh

Yiddish

Yoruba

---

Apologies for the formatting, not sure how to make it look nice in the comment.

A new bugfix update for the "Translate to English" toggle (which was functionally always set to on) should be available soon, it's just awaiting Play Store approval.

Whoa cool, so all the transcription is done locally? Have you done any perf monitoring around that on iOS?

I have been using the iOS built in speechTranscriber and it is... not great, was gonna use a whisper API but running it on device would be amazing if it isn't too heavy.

Hey, just spotted your comment.

Yes, everything is done locally, stick your phone in airplane mode if you want to be sure!

I'm not an Apple fan, but I have to say I've been testing it on an iPhone 15 and my god, the performance is insanely good, I was seriously blown away. I haven't dug into how much it impacts battery, but the transcription literally takes seconds for a minute of audio so it's not holding up your device.

The iOS version is built, ready to go, there's just some bug with my Apple account and it won't let me pay the £80 fee to signup (support ticket raised and waiting). As soon as that's sorted it'll be out on App Store for free as well.

Very surprised to hear the built in transcription is not great, anything specifically bad about it? The hardware is there.

You can download the desktop version from here (https://blazingbanana.com/apps/whistle/) if you want, still very much a WIP.

I really liked wisprflow on my mac but my daily driver is Manjaro KDE. I have stitched together a bash script that copies the transcription (right now I am using the Parakeet TDT 0.6B) to my clipboard. I would give this a try on linux when it becomes available.

Just checked out whisprflow, I must say that looks really nice, kudos to those devs. Shame there isn't a Linux / Android version.

I have added the auto-copy to clipboard functionality that will come with the next Android release and be included in all others. Adding a hotkey / quickbar button is on the roadmap for the desktop versions.

If you want to give the Linux version a shot, you can download it from here - https://downloads.formait.app/whistle/linux/WhistleDesktop-l... - I've just stuck it in the same R2 bucket as another app, as I've not sorted the proper pipeline out yet.

Would you be open to sharing your script? I run whisper.cpp in Linux through some stitched together scripts (https://news.ycombinator.com/item?id=44949314), but would be very curious to try Parakeet. I don't believe I can run it through whisper.cpp?

I looked at your script. I am doing basically the same thing just using onnx-asr (installed as a uv tool) with the parakeet model instead of the whisper-cli. Look Here: https://github.com/istupakov/onnx-asr

I just tried running this on a 30 minute meeting with some 10 people in. It got to the end, then just bailed without transcribing. I also did not get any errors or anything.

Really sorry about that, longer running audio (~10m+) is something I'm working on along with handling multiple speakers.

I've been focused on getting functional parity across all OS's since the Android release. This is very close to being done and I just need to reach the milestone of it being available on all platforms before I move forward.

Hopefully you will take another look when the next update is out.

Are you piping the audio buffers straight to the transcriber as they come in? Or capturing the entire recording and then processing it at the end?

I'm capturing the audio first, which was fine for short audio but ended up killing the process when it got too big. The updated fix is out now.

I did a complete overhaul to the pipeline so that it splits and processes at the end, this seems to have sorted it. I'm thinking about doing each transcription segment as it's coming in (with a bit of a buffer / overlap to keep context) much like the live transcription does, but for now performance is ok. Something I'll keep in mind once I've crossed some other things off the list.

@blazingbanana

We have a similar product in the construction space. Would love to talk to you about some of our challenges and possibly work together. Interested?

Very interesting, happy to discuss this privately.

Would you consider adding it F-Droid?

Yes absolutely! I'm a GrapheneOS user myself so understand not wanting to have to go through the play store if you can help it.

I believe you have to make the source code public (please correct me if I'm wrong). I'm more than happy to do so, I've used a whole bunch of open source stuff to build the app so it only seems fair, I just need to make it a bit less messy and something I don't mind being public.

Yes, not just public, but also licensed under a license that permits free redistribution, modification, etc. This is awesome!

nice app!

if I am talking in german the text is translating it to english. Didn't expect that

Thank you!

There was a bug causing the "translate to english" to be always enabled. This should work correctly and translate to your native language.

Will be in the next update (in a day or two).

Reply on news.ycombinator.com