Skip to main content

Paragraph splitting, auto-listing, and minor text corrections

Feature Request

Coming from Wisper Flow,. A few of the things that I really appreciated that I was hoping could be included in here are:

1) splitting up long messages into separate paragraphs automatically

2) resolving very minor each mistakes in an automated way. So for example if somebody was to say “3 o’clock, no wait, 4 o’clock”, it would be great for the 3 o’clock to not be there.

3) Adding an ability natively to create lists when somebody is listing something.

I think some of this is doable with the own customization settings, but it'd be really great if this was somewhat native. I found that when I was playing with tones that it would substantially change what I was actually saying at times, which wasn't ideal.

Overall, I really love this app. Amazing work to the Vowen team!

Posted by jolystep ·
Status: Completed7 comments

Log in to comment and vote

Comments7

  • jolystep

    •

    Mar 23

    Thank you! I did some testing on v 0.2.0 and I’m not getting the expected results (tested on Parakeet and OpenAI Large v3 Turbo). Maybe i’m doing something wrong though!

    Minor text corrections:

    Example 1: “Hey, can you meet at three o'clock? Actually never mind, sorry, four o'clock.”

    Expected 2: “Hey, can you meet at four o'clock?”

    Example 2: “Hey, does three, no, sorry, four o'clock work?”

    Expected 2: “Hey, does four o'clock work?”

    Auto-listing:
    Example 1: “Could we get some groceries today? How about the following? Get some Chicken, carrots, onions, broccoli”

    Expected 1: “Could we get some groceries today? How about the following? Get some:
    - Chicken
    - Carrots
    - Onion
    - Broccoli”

    Example 2: “Let's go ahead and address the following issues. One, when people are late to work and they're not doing anything about it, they should be messaging the group. Two, when people are eating lunch and they're not doing their dishes, that's creating a mess. And three, when people are not adding comments in their code, it makes the code harder to read.”

    Expected 2: “Let's go ahead and address the following issues:
    1) When people are late to work and they're not doing anything about it, they should be messaging the group.
    2) When people are eating lunch and they're not doing their dishes, that's creating a mess. 3) When people are not adding comments in their code, it makes the code harder to read.”


    Paragraph splitting:
    Example 1: “So just to give an example of how paragraph splitting wouldn't work, I'm kind of expecting that if I was to go on a tangent and talk about some things that eventually if the text got long enough it would split into paragraphs. So at this point I think the first paragraph should have already been split, perhaps, if the topic was long enough. Because the initial part of this paragraph is kind of introductory and the second part is kind of elucidating on the first part, so it seems like it would be different. Without putting a new line between every sentence, it feels that um this is not being done at all right now in Vowen. Really I think this would be a super awesome capability in Vowen and I would love to have that just so I can use it more when it comes to my interpersonal communication. For now everything is amazing and perfect about AI, coding and working with AI, no issues. But when it comes to just being able to speak and have perfect conversation with people in teams and whatnot, paragraph splitting, auto listing, and minor text corrections would be super helpful.”

    Expected 1: “So just to give an example of how paragraph splitting wouldn't work, I'm kind of expecting that if I was to go on a tangent and talk about some things that eventually if the text got long enough it would split into paragraphs. So at this point I think the first paragraph would split perhaps, because this initial part is kind of introductory and the next parts are kind of elucidating on the first part, so it seems like it would be different.

    Without putting a new line between every sentence, it feels that um this is not being done at all right now in Vowen. Really I think this would be a super awesome capability in Vowen and I would love to have that just so I can use it more when it comes to my interpersonal communication.

    For now everything is amazing and perfect about AI, coding and working with AI, no issues. But when it comes to just being able to speak and have perfect conversation with people in teams and whatnot, paragraph splitting, auto listing, and minor text corrections would be super helpful.”

    • Vowen

      Team•

      Mar 23

      Hi@jolystep the paragraph splitting, auto-listing, text corrections and other enhancements to the text are only done when “Enhance transcription with AI” setting is enabled. The base models themselves unfortunately do not offer these capabilities. I can suggest configuring “Gemini Flash 2.5” or “Groq” as they offer generous free API limits for daily usage.

      You can enable “Enhance with AI” setting from “Configure your AI” section of the app. Let us know if that helps

      • Xavier

        •

        Apr 13

        Hi, AI does not seem to be working for me lately (tried both groq and gemini) with custom instruction and can’t get the bullet or list. Would be good if it was by default, but understand this can be complicated.

        • Vowen

          Team•

          Apr 13

          Hey @Xavier , it’s likely Gemini / Groq is rate limited when you’d tried it. Do you see the “AI Enhanced” badge in the voice log, despite have “AI enhancement” setting turned on?

  • Vowen

    Team•

    Mar 22

    Hey @jolystep , we’ve improved our base prompt significantly to address these issues, so that all of these are handled properly. Thank you!

    • jolystep

      •

      Mar 23

      <3

  • Vowen

    Team•

    Mar 18

    Thanks for the feedback @jolystep

    Turning on “Enhance with AI” does all of these enhancements. Though, the tones mode is in “beta”, so there are some gaps to address there. But for “Enhance with AI”, it should not change what you are saying, it should just correct spellings, format it into lists, pick up auto-corrections and split up messages into separate paragraphs. The model we’ve found to work best are Groq Llama 70b and Gemini Flash 2.5 Lite. Could you tell us if you had “Enhance with AI” enabled or which model you’d configured?

    Since there is a lot of variation in how it can be said, we’ve found that it has to be done through LLM enhancement layer. We’ll brainstorm if it’s possible to do this natively.