Transcription tools that run locally?

Do you know of any tool to transcribe audios with sensitive information that works locally?

You could try to set up Whisper GitHub - openai/whisper: Robust Speech Recognition via Large-Scale Weak Supervision

1 Like

EDIT: removed, read thread title as “translation tool” lol.

As someone noted, https://voiceinput.futo.org/ is great.

Futo Voice if you are looking for an android app. You download one of three models - small, medium, or large - after that it works completely offline. The largest model is the most accurate but takes the longest to convert the speak to text. Runs great on a Pixel 6a but is terrible on my Samsung tablet.

1 Like

I second whisper. You can also check the app macwhisper if you have a Mac. Download the model and do the transcription offline.

audapolis was perfect for a specific workflow I had a few months back.

Edit: Activity on the project seems to have stalled. This looks like the upcoming replacement?

Is my PC able to run the large model? I can’t find the system requirements in the GitHub page: Processor: Intel Core i5-13500, 14C/20T 4.8 GHz LGA 1700
Motherboard: ASUS PRIME B760M-A WiFi socket LGA 1700
RAM: Lexar THOR OC DDR5, 32 GB (2x 16GB) 6,000 MT/s
SSD: Silicon Power UD90 2TB, PCIe 4.0 5,000 MB/s
Graphics Card: Sapphire Pure Radeon RX 7800 XT OC
Power Supply: Corsair RM750e, modular 750 watt.
I have 16 gb of VRAM which is 6 gb more than the VRAM usage but I don’t know if according to my other PC parts I can run it, probably yes since it cost me around 1500€ but just to be sure I’m not wasting my time

Yeah you’re fine

I have an older friend who is looking to dictate sensitive information to readable text without involving a third party (another person) or having it be online in any way. Is there a Speech to text software or device that can be used to relay his information without it being connected to anything? I understand just writing it down would be the safest bet. But in the essence of saving time, does anything like this exist? Or is there a way to make this possible?

I think FUTO voice might be what you are looking for

3 Likes

If they have the GPU power, pretty much all speech-to-text huggingface models can do this.

1 Like

Thank you for all the responses thus far. He is not exactly what you would call tech saavy haha I know he is using Apple products, Phone and PC. I think he is looking more towards relaying book type information of a larger quantity.

For Apple platforms you could try Aiko. It uses OpenAI’s Whisper model just like FUTO voice and works entirely offline on both iOS and macOS. It uses the small and medium models on iOS and the large model on macOS, so I assume it would be more accurate on desktop but YMMV.

3 Likes

This looks like exactly what he’s looking for. I will research and check it out. Thanks!

Opening the conversation up.

I have no accessibility requirements personally, but am I big user of Speech-To-Text. In my experience, it’s faster than keyboard typing (and even swiping) and an overall more pleasant/natural experience.

I got addicted to it w/ Google and tried a few alternatives before I landed on Whisper IME. STT was the one thing holding me back from fully embracing the FOSS keyboards, but this app is what led to deleting GBoard. It can be used offline permanently once you download the necessary model and is very accurate. I highly recommend it as I was surprised just how accurate the dictation was

2 Likes

I’ve just started using Whisper+ (From F-Droid. I don’t know if it’s in the Play store.), and it seems great on a good phone. I was using Sayboard (also from F-Droid) before, and it’s not as good, but I’d reccomend it for phones that can’t run Whisper-based speech-to-text well, especially English-speakers. I was happy with it, apart from its inability to transcribe my Spanish. I only switched because I updated my Phone and it won’t run on Android 14, but Whisper+ is much better on a phone that can run it well. Sayboard definitely works offline, and so does Whisper+. I found a Whisper (without the “+”) which installs on older Android versions and looks very similar, but it needs a download nearly twice as big and still doesn’t run well at all on a slow or low-RAM phone.

1 Like

Whisper+ is really nice with HeliBoard.

What would you recommend for Windows?

If you can navigate your way with a few small CLI commands, you could install pretty much anything from Github (OS is irrelevant in that case). :+1:t2: