[Generative AI] DeepVoice - Text To Voice

I am liking this at the moment and of course looking forward to all future updates and additional voices and languages. I have only been using this since yesterday evening so I will get back with more suggestions the more I get into it.

- Equalizer needs to have the ability to entre values and not just sliders. Also we need to be able to save and load the settings.

- Status debugs how many characters used but it should also debug how many characters are left. This should also rather be a popup window and also verification should also be a popup window if verification successful or not.

- Voices should be separated into male, female, boys and girls. Then also later on down the road when you have more voices, have categories like famous, adventure (pirates, villains.)

And languages or dialects should be easy to recognise like if it is English irish, scothish, welsh, and so on.

- Emotions like, he said hurriedly (should not be added to the audio result and should not count as characters used.)

- If the same text is entered it should be possible to experiment with different generation settings without burning your character count as it is obvious that you are trying to get it write first before committing.

- Why do we need to use \” and end “\ Can A.I not work it out for itself that we want it to just generate the text we write?

- The example voices need to be longer.


My question now.

What about pirates? Just think about how they sound in all the films. Female and male pirates would be great.

 

Hey there,
I tried to activate your asset but in Unity 2023.1.14f1 I get the following error message:

InvalidOperationException: Insecure connection not allowed
UnityEngine.Networking.UnityWebRequest.SendWebRequest () (at <1ed9028a981141979c554601915576ad>:0)
AiKodexDeepVoice.CanvasController+<Verify>d__40.MoveNext () (at Assets/DeepVoice/Demo/Resources/CanvasController.cs:210)
UnityEngine.SetupCoroutine.InvokeMoveNext (System.Collections.IEnumerator enumerator, System.IntPtr returnValueAddress) (at <cd14705e8f094afcb566740ab5e8c02d>:0)
UnityEngine.MonoBehaviour:StartCoroutine(IEnumerator)
AiKodexDeepVoice.CanvasController:VerifyInvoiceButton() (at Assets/DeepVoice/Demo/Resources/CanvasController.cs:108)
UnityEngine.EventSystems.EventSystem:Update() (at ./Library/PackageCache/com.unity.ugui@1.0.0/Runtime/EventSystem/Event

I pretty much have an empty Unity project and a new scene with only 2 other plugins imported.
Any suggestions?

 

Hello,

You require to change a few settings for this error. You can either fix this by opening the DeepVoice editor (Window > DeepVoice) (which includes an auto switch to development mode HTTP request) or manually by going over to the project settings and enabling insecure connection during development mode only. Just for clarification, this isn’t necessarily an insecure connection, it’s an HTTP request without the SSL certificate. Please reach out to us on info@aikodex.com if you are still having issues so we can look into your issue.

 

Is it difficult to answer this question? it seems like this is an awkward question for you:

 

Hello Fanat91,

Sorry for the delay in our response. We have been working on an offline solution since the release of the asset so that users would not have to rely on cloud. At the moment there are several challenges to this, but with the advancement of ONNX open neural network exchange models ability to process bigger models with a faster inference time locally as well as the hardware becoming more capable (GPUs receiving more CUDA cores) it will make it easier to port the models to work offline with the same capabilities. For now, running the service on a cloud is the only solution. We plan to maintain cloud services for DeepVoice AI till the time we do not ship an offline model that is compatible with every device with the DeepVoice AI asset.

 

Hey there,
I tried to generate a voice but since yesterday (no problem before and no change) i have this error in all deep voice model (unity 2022.3.7f1) :

 

any plan for a small sale in near future? otherwise ill just buy full price..

 

Hello eagleeyez,

Thank you for outlining these points. We are working on these points and will hopefully release them in the next few updates.

As for the syntax, we will try and figure a way around it, perhaps wrap it in another layer of string operations.

Unfortunately, the characters still count while experimenting with different generation settings for the simple reason that it is another request that the server has to compute.

With time, better models shall come along which should get the emotions correct more accurately.

For pirates, we will see if we can synthetically generate a pirate sound.

 

Hello TheQuantumWanderer,

Please send us an email at info@aikodex.com with more information such as the prompt you are using, the asset version and the generation settings.

 

Hello unitygnoob008,

We as publishers do not come to know of Unity sales schedules till shortly before the assets are picked and put on sale.

 

Hi, hope development is going well.
Did you have any progress on adding older character voices? Old, soft, and warm would be perfect Fireside story time, not grumpy old man. Though in the future I'm sure they'll be a need for that somewhere. haha

 

i got this error now i cant use the trim and equalizer UnauthorizedAccessException: Access to the path "C:\Users\USER\FOLDERWHERETHEPROJECTIS\THEPROJECT\Assets\DeepVoice\Voices\Temp_data\TempTrim.wav" is denied.

 

Hello Cooky6,
Sure, in the next update, we will surely add older voices.

 

Hello KenzoGames38,

This error seems to be either related to the permission settings or file creation.

The reasons could be as follows:
- The caller does not have the required permission. (You can try running the project as an administrator)
- Path specified is invalid. (Please ensure that the file name field is not blank as this will cause an error)

Please let us know if these fixed the issues.

 

tried both and it still don't work also the problem only happened since the last update EDIT: i manage to fix the issue by deleting the temp folder, will you add Japanese accent voice in the future ?

 

Just picked this up. Was wondering 2 things.
1. How do I add longer pauses between sentences. I've tried combining period, dot-dot-dot and new paragraphs. Can I get longer pauses?
2. How do I translate to different languages? I am using the Multi voice and can see what is supported but don't know how to switch languages.

 

Greetings, I was thinking of buying this assets for an important project. I just wanted to know if it was possible to use a personalized voice (with a training of course), such as mine or a friend (or if it is in the roadmap)

 

The following is just intended as a suggestion/healthy criticism. Don't hate me too much, lol.

I'm considering adding text-2-speech to my project, so I'm reviewing assets like yours. And I like what I hear from the demos; the implementation seems solid enough. But the way you're managing the quota is offputting, to be honest.

Ofc, I appreciate the fact you've been increasing it. That's a pro. But the thing it is capped at a given number, for all users, regardless if they've used it or not, it's disappointing.

Making games, especially on your own, is not a homogeneous process. I may very well spend a whole month adding engine features, and not touching dialogues at all, only to focus on that in the second month, and find myself with the reset quota. Not fair.

Also, the 200-character limitation per audio is pretty crazy if you ask me. Especially when the 'emotion tags' aren't actually tags but descriptions. I don't know your reasons. There must be a reason, I'm sure. But if I were you I'd focus on solving that, because forcing users to manually join multiple audios to form a full-character interaction longer than a sentence is also very offputting.

Finally, you should add a way to buy more quota on demand, without needing to switch invoice numbers, and it would be nice if you could offer a very limited demo, just to review the voices and the UI.

For all these reasons, even though I think yours is the best-sounding solution I've found here, I can't take the plunge.

 

I have been messing around with this asset and not getting much traction. The voices are nice but the control you get over voice pauses and intonations isn't that great. I was hoping this could supplant what I'm currently using now but it looks like Amazon Polly ( https://aws.amazon.com/polly/ ) is still the best option. You get much better control over pauses between sentences, breathing, etc. when using SSML.

 

By the way, still waiting for answers to my questions above. #116

 

Hello KenzoGames38,

Thank you for sharing with us the fix to your problem. We will try fixing the core issue. Perhaps we can change using hard coded paths and assuming directories. This may take some time.

As for Japanese voices, you can switch to the Multi model and try out different voices.

 

dev, is it possible to add more tools to warp/adjust/modulate voices generated? its not necessary, but just a suggestion to make the tool stand out even more

 

Hello jnbbender,,

A longer pause between sentences can be achieved by (...) or (-) or new lines. However, if you are not having any luck with that, you can use the built-in audio trimmer to separate the two sentences into two different files. You can play these files during runtime with a gap of any desired length if you are using the timeline to create a cutscene or with a script in-game. Although this may not be the most elegant solution, it allows you to customize the pause to your exact preference. Another option is to digitally manipulate the clip more effectively in Audacity, a free audio toolkit.

As for generating audio in a different language, please select the Multi model and enter your text in that language. The AI will autodetect the language and generate voices accordingly.

 

Hello Remer,

At this time, unfortunately, it is not possible. We cannot provide an estimate for when we will be able to introduce personal voice training to the asset. We will inform the community through the forums once we are certain that it is feasible.

 

Hello unitygnoob008,

Thank you for this idea. Could you please elaborate more on what you mean when you mention warping, adjusting, or modulating voices? Do you have any examples that could help us better understand your suggestion?

 

Yes,

tools such as audacity feature plugins to change pitch, tone, echo, amplify, etc. of sound clips.

Your tool features a nice in editor clip extension, but I also saw you had some pitch adjustment already.

You can run audio through sound editor, or modulator, should you wish, that can change the way input audio is output once processed similarly based on a preset... but I will admit, I am not overly familiar with any of this, and may just sound like I am talking from thin air.

Warping would be distorting the voices to sound differently, here is what I am saying:

In that game the voices are the same, but adjusted to have differences (pitch, amplification, bass, speed, etc).
However, I was thinking you could potentially run generated voices through pre-set modifiers which can take the audio, and reprocess it with settings that make it sound a certain way.

Anyway, it was just some thinking about it.

 

Assets\DeepVoice\Editor\Scripts\DeepVoiceEditor.cs(7,13): error CS0234: The type or namespace name 'EditorCoroutines' does not exist in the namespace 'Unity' (are you missing an assembly reference?)

and i open script
it error on this
using Unity.EditorCoroutines.Editor;

 

To resolve the issue with the missing Editor Coroutines package, please go to the package manager, search for Editor Coroutines in the Unity Registry and click on install. The package is offered for free by Unity in 2020.x and is inbuilt in versions 2021.x and above in Unity SRPs.

 

[Announcement]

Automatic Quota Reset to 60,000 characters. [Quota changes]

60,000 characters allotted for the period 1-10-2023 to 31-10-2023.

 

I would absolutely love it if there was an option to have different versions of English accents.
So you could select British/Spanish/French/German/etc as an accent. This would give characters using the voices a lot more personality options!

Same for age!

Edit: I suppose the Standard voices do this to an extend. But finding the one that is right based on names makes it harder than it has to be. And the mono/multi voices are a lot better as well!

 

Hello Boaike,

Thank you for your suggestion.

Accents can be induced by misspelling words for mono and multi models. For languages in a different script, you can use the script to write the language but have English words. An example would be:

This is a brilliant method which works to some extent proposed by Kyntaxa. Perhaps you can try it out for yourself.

Other than that, yes, standard voices can simulate a few accents, however, the quality is definitely lacking in that model.

 

First and foremost, I want to express my gratitude for creating this incredible asset, but while it does have potential, there are some areas for improvement that need to be addressed.

the asset's reliance on a server for text-to-speech conversion has its drawbacks, mainly the character quota limitation. It can be restrictive, and it's frustrating that the character quota is deducted even when an error occurs, and no speech is generated, you only get a 1KB MP3 file that can't be opened.

I would like to suggest the following enhancements, either:


Remove the quota limitation or make it work offline: The character quota limitation needs reevaluation. Either remove it entirely like in your other asset "AIFY", or make it work offline so we don't have a characters quota limitation (Some of us have powerful GPUs and can handle offline processing on our computers), Developers should not feel penalized for unsuccessful attempts or errors, especially when they don't receive any usable audio.

or

Refund Quota for Regeneration: It would be a significant improvement if the character quota is not deducted when regenerating the exact same phrase with the same words. This feature would allow users to fine-tune the generated speech to achieve the desired feeling without worrying about wasting character credits.



I believe by addressing these issues, the developer can significantly enhance the asset's value and make it an essential tool for Unity developers.

I hope the developer takes these suggestions into consideration.

 

Hi AiKodex!
Is it possible to put the correct stress on a word?

 

Hello !

After generating an audio, when I want to use "Audio Trimmer" etc. I get the following error:


The dependency is installed.
Do you know how to fix this problem?

Thanks in advance !

 

Hello MagicRS,

This error usually occurs if you select a file that has been already changed or modified since you last used the Trimmer.

The standard procedure is to click on the clip you want to trim, select Active Clip, set the markers, give the file a name and click on Save Trimmed Audio.

Please let us know if you've tried the above and are experiencing the same issue.

 

Hello Jaco93,

Thank you for your suggestions.

We have been working on an offline solution since the release of the asset so that users would not have to rely on cloud. At the moment there are several challenges to this, but with the advancement of ONNX open neural network exchange models ability to process bigger models with a faster inference time locally as well as the hardware becoming more capable (GPUs receiving more CUDA cores) it will make it easier to port the models to work offline with the same capabilities. For now, running the service on a cloud is the only solution. We plan to maintain cloud services for DeepVoice AI till the time we do not ship an offline model that is compatible with every device with the DeepVoice AI asset.

We deduct the number of characters according to the number of bytes and the computation required to generate a response from the server. If it was possible to ship an asset with large TTS model and assumed the user's machine to be capable enough to run it within reasonable time and output a response, we would deliver an offline system with an unrestricted and unlimited access to the model.

 

I wonder why the language "Bengali" has not been added in spite of it being the 6th most spoken language in the world.

Here is your reference

The 10 Most Spoken Languages In The World

 

Hello
I purchased your asset yesterday.

I Had to install coroutine editor ( found the answer on the forum ) and then tried to verify invoice number .. but to no avail!



I bought it and still haven't been able to test it!!
Best regards,
Eugene

 

Hello Eugene,

This is concerning. Please write to us on info@aikodex.com and hopefully we can get it working for you in no time.

 

This has stopped working this evening. It was working fine but now all I get is...
"There was an error in generating the voice. Please check your invoice/order number and try again or check the documentation for more information."

Invoice is fine and verifies correctly.

 

Hello Mulbin,

If the invoice request is successful, it is very likely that the problem lies in the string being inputted. Please ensure you remove all special formatting (please paste the text in a notepad before) or try and type out the text (for testing purposes). If it still fails please contact us on info@aikodex.com

Another pointer: Please do not press enter after inputting the text as that adds an illegal character at the end of the request which results in a bad request error.

 

Can i use this package in WebGL unity?

 

Unfortunately, we designed this asset to be only used during development.
A few developers have however managed to get it working in Android iOS and webGL builds.

 

Is there voice cloning?

 

Hello DerDicke,

As of now, DeepVoice does not offer voice cloning. We’re still working on setting up this feature.

 

what is the catch with this? I am limited or something. It says some 60k lines or something per month or once it reaches the limit we can't use it anymore? I need some life to my characters' dialogues and maybe cut scenes.

 

[Announcement]

Automatic Quota Reset to 60,000 characters. [Quota changes]

60,000 characters allotted for the period 1-11-2023 to 30-11-2023.

 

[Announcement]

There is a technical snag in the Mono/Multi voices currently.
The Neural and Standard voices are functional.

We hope to resume services by UTC 16:00 Tuesday, 7 November 2023.

Edit: The services should be fully functional now.

Thank you for your patience.

 

Hi!
Is there a demo page where we can hear previews of all the voices?
Thanks!

 

Hi
Does Hebrew language support?