Feature Request: AI Text-to-Speech Device for UEFN
Hello Epic Team,
First of all, I want to say that the new Persona / Conversations system is an amazing addition to UEFN. It opens the door to much more immersive storytelling and interactive NPCs.
I would like to suggest a feature that builds on the same technology.
The idea
Create a Text-to-Speech (Narration) Device that allows creators to enter a predefined text and have it spoken using the same AI voice technology as the Persona Device.
Instead of requiring an NPC conversation, creators could simply trigger a text to be read aloud through Verse or device events.
Why this would be useful
This would unlock many new possibilities, including:
Audio logs
Story narration
Radio messages
Computer or AI announcements
Museum or exploration guides
Quest descriptions
Environmental storytelling
Automatic localization
The biggest advantage would be localization.
Today, if creators want voiced content, they have to record and import separate audio files for every supported language. This quickly becomes impossible because of storage limitations and the amount of work required.
With AI-generated speech, the device could automatically generate the narration in the player’s selected language, just like the Conversations system already supports multiple languages.
This would:
dramatically reduce project size,
eliminate the need to store dozens of WAV files,
make updates much easier (only the text needs to be edited),
allow even small creators to offer fully localized voice acting.
Why it fits UEFN
Since the Conversations system already combines AI text generation with AI voice synthesis, this feature would reuse the same technology while enabling many non-conversational storytelling experiences.
I believe this would become one of the most valuable tools for narrative creators and adventure maps.
Thank you for considering this suggestion, and thank you for continuing to improve UEFN!
That is actually my next personal project in the queue, developing a live TTS that can be used locally, completely offline, and independently of remote computing services. Let’s see what’s possible.
However, it would be nice if Epic Games would offer a remote service, that provides a wide variety of voices and with different moods for live TTS.
Really loved this idea. I also think a standalone aI text to Speech device would be useful well beyond NPC conversations. Being able to trigger narrated text from Verse or events would make things like tutorials, announcements, audio logs, and dynamic quest updates much easier to build. Also, the automatic localization aspect is probably the biggest advantage. Instead of maintaining voice recordings for every language, creators could simply update the text and let the engine generate speech for each player’s language. That would save a lot of storage space and make narrative heavy UEFN projects much easier to maintain.
Thanks for the support! I’m glad you see the same potential. Hopefully Epic considers building something like this on top of the existing Conversations technology. If you like the idea, an upvote would really help give it more visibility
Here my own proof of concept. However, there is one thing to keep in mind— TTS isn’t worth it if there isn’t much spoken audio, since the file size of the NN model exceeds that of the audio files. Also, the text-to-speech conversion takes a few milliseconds. But that might be fixed with something like a pre-load.
Right now the speech is still a bit emotionless, but that can be adjusted, including voice cloning.
I used the pre-built NN model of Kokoro TTS, which is currently limited exclusively to English voices. But i also implemented Piper TTS for a German voice. Piper TTS offers a whole list of neural network models for voices in various languages, but their quality isn’t that good.
There are certainly other text-to-speech models available, but you need to check their licensing terms. Not all models can be used for free or be used in your own commercial products.
Maybe someday I’ll be able to train my own TTS model, complete with voice tones and voice emotions, as well as voice cloning. Unfortunately, I don’t have the computing power to get that done in a reasonable amount of time, and have no idea where to get the training data either.