Hello Transcribe

I'd been working on transcribing recorded meetings on mbp 16 / 2019 Intel, and the large-v2 or -v3 models could take 14hrs on a 3hr recording. I looked for an alternative, saw the comments about the Apple silicon / Metal improvement, dropped that sucker on my iPad Pro 11 (4thGen / M2), and it ran that same process in under 2 hours. Amazing. Quality on 'medium' model a little gappy sometimes, but the recording wasn't perfect either. Very impressed - just added another monster solution to the swiss army knife iPad.

Please fix them.

Broke

What I would love to see if this could keep the audio file attached (.m4a from voice memos). For both reference and posterity in case I need to double check or verify transcription audio for any errors in the transcription. Other than that, I love the whole app and accuracy. Kill subscription/server models and give us the option for offline and private transcription! Nice work!

Fundamentally, this is a pretty good implementation of Whisper on iOS. For longer transcriptions, it appears to just chunk the audio into 30 second segments, which can degrade the quality of the transcription, including unexpected line breaks and words that overlap two chunks not being great. I wish there were a few more options. One simple one would be a one-click button that allows you to re-process a finished transcription as a single complete audio file. Alternatively, I would like to set it up so that the real time preview is done using one model of my choice (maybe tiny, maybe base), and then after the transcription is complete, it would automatically process the complete audio with another model of my choice (maybe small or medium). I go back and forth between this one and WhisperBoard. They’re both good in their own ways, but transcription quality is critical, so I rate WhisperBoard higher since it doesn’t seem to suffer the same oddities that are the result of Hello Transcribe attempting to do real-time transcription.

I’ve been waiting too long for apple to bring AI to voice memos but they haven’t done anything about it yet. I like that it uses whisper locally but would also have the option to just record audio and then let it transcribe overnight when I’m not using the phone (this could also enable us to use larger models since the transcription doesn’t need to be real-time); this in turn should allow me to search later through my recordings and when I tap on a word from the search result it should move the audio player cursor to that position, please implement something like this! Thanks, it’s almost there

This aop has no access to Medium or Large Models.

This app is EXACTLY what I’ve been looking for! As someone who relies on journaling/freewriting for everything—from internal reflections to business brainstorming—I’ve wished for an effective voice-to-text solution. Other apps either lacked live dictation or were really inaccurate (looking at you, Siri). Hello Transcribe, however, has been a lifesaver! Now, I can freewrite on-the-go, while walking or driving or washing dishes. The hours I’m reclaiming, and the creative surge I’m experiencing, are simply monumental. And feeding these transcriptions into ChatGPT has been incredible - giving me summaries, turning rambling thoughts into article drafts, emails, action items (and even app reviews 😉). And surprisingly, it even runs in near-real-time on my old iPhone X! I do have 2 suggestions though: the ‘clear’ button is dangerously close to ‘pause’. Wish there was a confirmation prompt. And an option to keep the screen awake during recording would be really nice. But even still, this app, especially at its price point, is perfect to me. Thanks for creating this Ben!

Not sure how Ben did it, but this implementation of Whisper is just as fast as any other I've used (such as MacWhisperer) on my Macbook Pro M1 Pro -- and yet it ONLY is using up RAM when actively doing the listening/transcribing infernece. I'm assuming that has something to do with CoreML and the 20-minute one-time initialization that took place when i first installed the ~600mb Eng Small Model. A couple things that would be amazing to have in the future: - a menu-bar version for quick/easy access to transcribing (or even something like the ability to right click in a text field and select an option for transcribing voice to text, if that's possible). - MacWhisper has an option to clean up the random line breaks and spacing that happens during the transcription, would love to see Hello Transcribe get the same ability. - Ability to use the Macbook's system audio to transcribe any audio source

This app is great! I do most all my computing on the App Store-- both for work and for personal use. For my profession, I also do a lot of transcription but need a transcription app that allows me to keep my transcription on my device and not send it over the internet-- this app does all of that! I tried different apps in the past but none achieved enough recognition and kept my information on my device (Apple's dictate is okay but not so accurate when not transmitting information to the cloud). The app is also free and the developer shows the user in the settings good information about what is going on with the app and customizations. I opted to also purchase the "Pro" version which gets you a bigger speech model (more accurate) for a one time fee of approximately $7 (seems quite reasonable)-- rather than other Whisper apps often asking for $99/year (especially considering the underlying Whisper engine is open source and free). Looking forward to the developer continuing to develop this app. Developer might want to consider a catchier app name to attract even more interest on the App Store.