I’m trying to import kokoro-82M (TTS) onnx model but I get these errors (I’m using the latest Sentis 2.1.1):
ConcatFromSequence not supported
Loop not supported
SequenceEmpty not supported
SplitToSequence not supported
when are these going to get supported?
I am not in the position to promise anything. We will take a look at the kokoro model and the missing operators.
Meanwhile, here is the list of supported ONNX operators .
This model is great. I would like to use it in my games with sentis. Please support it.
Another vote for Kokoro please, searching for “kokoro Unity” brought me here Thanks for all your hard work!
We really need more TTS solution so I’m adding my vote
Mchat
February 20, 2025, 3:54pm
6
Yes, this would be fantastic. Kokoro seems like one of the best quality and small TTS. Would be fantastic to get it to narrate some parts of the game.
Etonix
February 27, 2025, 10:14pm
7
Looking forward to being able to use Kokoro!
I would also like to see Kokoro in Unity.
I’ve tried it out with the new Unity Inference Engine, but unfortunately, it’s still not supported due to missing operators:
Are there any updates about supporting these operators?
We have developed an alternative to Kokoro called Lingotion Thespeon - it does not use the unsupported operators and thus works with the current version of Inference Engine. If you are interested in trying it out, check out our github repo!
NiklasP
February 24, 2026, 8:54am
11
Unity is now officially supporting Kokoro TTS. They have added an example here:
# Text-to-Speech Sample
Interactive interface powered by the Kokoro Text-To-Speech model running locally in Unity using Sentis.

## Runtime Inference
To power this experience we leverage the [Kokoro-82M-v1.0-ONNX](https://huggingface.co/onnx-community/Kokoro-82M-v1.0-ONNX) model, a high-quality text-to-speech model.
The system processes text inputs through:
- Text tokenization and grapheme-to-phoneme conversion using our C# implementation of Misaki for English
- Neural voice synthesis using the Kokoro ONNX model
- Real-time audio generation with multiple voice options
- Configurable speech speed and voice selection
We use this to create a seamless text-to-speech experience with natural-sounding voices.
## Features
This file has been truncated. show original
Interesting, it looks like those operators still aren’t supported: Supported ONNX operators | Sentis | 2.5.0