Note about version: Changed the version name to 1.2.1 from 5.2.0? Yes! xD.
Strange, but I was bothered by the old bad version management I had and decided to force a fix by returning the name to 1.2.1
There are two packaged versions (for UE5.0 and UE5.1) in the release page, to allow using before Marketplace approval without the need of compiling the plugin’ code: AzSpeech_UE50_PACKAGED.zip and AzSpeech_UE51_PACKAGED.zip. You can put it into your Project’s Plugins folder.
Instead using AsyncTasks directly inside the plugin’s UCLASSES, I’ve created separated Runnables to manage the recognition/synthesis work in NonBlocking threads and send the informations to the UCLASSES to call the delegates broadcasts.
Changes
Fix Viseme data broadcast delay
Fix cancellation log crash
Fix .wav file loading (In .wav File to Text)
Rename some functions & adjust some comments (Added core redirects)
Fix delegates and add a check to ensure that the plugin will broadcast only in game thread to avoid crashes in widgets calls
Added more logs & Debugging prints
Removed boolean bIsContinuous from recognizers, now they’re using only continuous: Continuous tasks are better to manage in the runnables, allow task cancellation/stop and work similar to the single recognition (if you stop talking, the recognition will stop too)
Fix thread blocking/game freezing when a task is activated/deactivated
Added new delegates: Synthesis Started and Recognition Started
Set the plugin settings to accept 10 candidate languages
Notes:
More information about recognition map is coming.
AzSpeech.zip is working in both UE4.27, UE5.0 and UE5.1
The function is now a Async Task and run in a background thread
Now this function will try to check the exact word instead only checking if the input string contains a substring by using String Delimiters to validate the word range
Fix delegates and add a check to ensure that the plugin will broadcast only in game thread to avoid crashes in widgets calls
Added more logs & debugging prints
Removed boolean bIsContinuous from recognizers, now they’re using only continuous: Continuous tasks are better to manage in the runnables, allow task cancellation/stop and work similar to the single recognition (if you stop talking, the recognition will stop too)
Fix thread blocking/game freezing when a task is activated/deactivated
Added new delegates: Synthesis Started and Recognition Started
Set the plugin settings to accept 10 candidate languages
Set some functions from UAzSpeechHelper as BlueprintCallable instead BlueprintPure
Hey, thank you for mantaining this excelent plugin! I am using the 1.3.1 version in Unreal 4.27 and I am trying to add auto candidate languages but nothing happens when I click to add an item. The only item that exists in the auto candidade languages array is the default language ID. Also, when using a speech to text node with an enable continuous recognition node right after that (on the first output exec pin - above Async Task), it causes a crash to the engine.
Pushed an new experimental branch, trying support for Linux, MacOS, iOS and HoloLens.
Unfortunately, I can’t guarantee that it will work, as I don’t have the equipment/environments to test these systems. But I hope it serves as an alternative to try and add custom implementations.
This branch has all the latest plugin updates and attempts to support all mentioned platforms.
Thanks for this amazing plugin, epic should hire you! I recently added it to my project and it works fine on windows. But not on steam deck it seems, no sound. I’m using the TTS function for voice. Any ideas what could be the cause? (the game: FlipScapes
Thanks!
Working with both UE4.27, UE5.0 and UE5.1
