Hi, I got my pocket yesterday and have been testing it out. It adds way, way too many speakers.
As an example: this morning I went to the doctor. I am a male. Doctor is female. Two people in the room.
Almost every single block of text is labeled Speaker 1… all the way to speaker 5. However, there are blocks of text that were 100% both of us. Like, I said something and she replied. And it combined that into Speaker 4.
I used the voice training and it’s still just labeling people.
Is there something else I can do?
This also happened in the two work meetings I recorded (two people, labeled as many more and very much confused people. For example, I’m listed as speakers 1, 2, and 4. He’s also got some parts as speakers 2 and 4.)
Speaker recognition is a tough one - even the top competitor out there, I have the device, doesn’t do a great job with it. I do believe the over time this will improve.
Is there a way to manually correct mistakes? I can’t figure it out, if there is… I’d like to be able to edit the transcript to correct these mistakes. Even some words that were mis-transcribed, it seems there is no way to edit the transcript?
So if it can’t really be done well right now why go through the whole voice recognition stuff? The fact that I did that made me expect that it would recognize my voice, I’ve never had that experience ration for something like otter.
I don’t know if it’s just me, but the more I use this, I feel like the transcription and speaker recognition are wildly inaccurate - in some cases, borderline unusable. Now, I don’t mind accepting that it’s new tech and will take time to improve, but in the meantime, I wish I had the functionality to manually make corrections in critical places that I decide to do so. I wouldn’t mind doing this either, again with the understanding that the AI tech is new and hopefully will improve with time.
Y’all gotta remember it’s a new product. We crowd funded the idea and product and it just launched. It’ll get better. They just released the website where we can submit our ideas for features and vote on em last week.
They’ll get to it! They’re responsive in here and have first time users to support plus features plus tweaks and bugs to figure out. They’ll get to the voice recognition and other stuff!!
It’s wild that the transcripts cannot be edited. Just like you described, if the transcript is wrong, then everything else derived from it would be wrong.
We shouldn’t have to request this as an additional feature. It should have been a core feature from the start. Every other AI recorder that I use, whether software or hardware, allows for transcripts to be edited.
I know this might be classified as negative, constructive feedback; we’ll see whether I get silenced or not, but seriously, this is unacceptable.
@p_odoi No worries - this still counts as constructive feedback, and we appreciate you and everyone else taking the time to share it. I am sure @Akshay will jump in when he gets a chance. The team is going through hundreds of requests right now.
Just a small note, it always helps when feedback is shared in a calm, solution-focused way. We’re all trying to make Pocket better together, and tone really does go a long way in keeping discussions productive.
@pete-w I appreciate your taking the time to respond. I hope you and the rest of the team understand that every critique is coming from a good place. We believe in the product, and we want it to get better. I still believe that the Pocket can be the one device to rule them all. I’m confident that your team will take all the features we’re voting for and implement them in upgrades in the very near future. Personally, I don’t want to have to carry multiple AI recorders with me simply for redundancy. I hope (and expect) my Pocket device to be the all-encompassing and ONLY device I will ever need.
@p_odoi yes, we completely understand that. Constant improvements are being deployed, and the team’s been working overtime nonstop to keep making Pocket better every day. While we can’t tackle everything at once, every comment is read and if something’s important, we’ll do our best to address it as soon as possible.
Is there a way to ‘introduce’ speakers? For example, the first time a speaker talks, they could say “Hi, this is Randy” and it would then match your voice with your name?
