Since the last update, my recordings which are just ideas and not meetings are listing every word as a different speaker. Is there some setting I’m missing? I tried merging speakers to see if I could get the transcript all in one post but that didn’t work.
1 Like
Hey @Xflisa speaker diarization is not available on the free plan and recording over 3 hours do let us know if this is not the case
This was probably a 15 second recording. There was only one speaker, me. Are you telling me I have to pay 20 dollars a month to get a transcript that isn’t chopped up like I showed in the screenshot?
1 Like
hey @Xflisa there is no speaker identification or splitting on the free plan on any recording length sorry
I don’t think you are understanding the question. Did you look at the screenshot? There’s only one speeaker, me. I don’t care if it recognizes me. I do care that it thinks every word I speak is a different speaker. I’m not recording meetings or long notes. If it splits a 15 second recording by one person into 20 people, that’s a huge problem. If I record something only 5 minutes it makes this product unusable.
The problem is new. I’ve been using it for over a month. Yesterday is the first time it created a transcript like this. It is doing this on every single recording.
Unusable. I was on the fence about paying for pro but I’m definitely a no if this isn’t fixed or someone explains a setting I’m missing.
If anyone else has read this far, are you having trouble or is it just me?
Thanks.
@Xflisa thanks for the detail. The transcription providers on free and Pro are different. We’ll look into why this started showing up, but Pro uses a higher-quality pipeline where you wouldn’t hit this. On free we can’t merge the segments above into one turn because we don’t have speaker info per segment to know it’s all you.
I have noticed the same frustration on my Pocket transcription too. A 10 minute conversation generated 78 pages of dialogue with breaking sentences into fragments.
sincerely frustrated,
Peaceful greetings from NE Ohio near the southern shore of the Great Swamp Erie.
Tell the AI to NOT identify speakers. Problem solved.
Here’s my customization entry for the AI’s instructions for my needs:
Create summary as paragraphs only. Do not break up into categories. Do not identify the Speakers. Do not use possessive pronouns.
1 Like
I’m very happy you solved your problem. Mine, however, is not solved by your declaration. The issue is not the summary, it’s the transcript. AI summaries are not practical or desirable for how I use this recorder.
You also didn’t explain where to tell AI to do anything. Where in the settings is this option? Is it in the free version? I see no setting with an option to tell the AI anything.
Didn’t know you were unaware of Templates. Thought your issue was fine tuning yours.
Try experimenting a bit.
In your device’s HPckt app, go to SETTINGS> SUMMARY>THEMES> then scroll down and review some templates and then go back to your recordings and choose REGENERATE and choose one of these premade themes
Or in the THEMEs area, choose CREATE NEW TEMPLATE.
This is where I was referring to “Telling the AI” in my previous post.
You can tell it like I said before about the speakers. Maybe put a sentence in there about make everything bullet points. Assuming you want all these random things you’re talking about as bullet points. Try coming up with different things. It’s similar to when you call your bank or some other company by phone and you have to deal with the automated assistant until you get to a real human. You have to give general sentences to get the point across.
I get you want the transcript aspect, but these templates might help clean up what you’re verbally saying instead of just being a transcript capturing every time you say Umm and uhhh, etc. it’s sort of the whole point of HeyPocket in my opinion to clean up and organize your words.
I don’t use the transcripts so I can’t help you, I’m sorry. But thought I’d give this last try to help with another option you may find is better once you get the hang of it.