Error generating speech: Failed to execute 'endOfStream' on 'MediaSource': The 'updating' attribute is true on one or more of this MediaSource's SourceBuffers.

[szaronko]

Describe the bug
Can't generate on any browser

Screenshots or console output
when generating on chrome I get this error.
Error generating speech: Failed to execute 'endOfStream' on 'MediaSource': The 'updating' attribute is true on one or more of this MediaSource's SourceBuffers

Once I waited for a while and it worked once, but then stopped working again, even though I didn't change anything. I played with all the options and nothing worked. **Branch / Deployment used** kokoro-fastapi-gpu:v0.2.2

Operating System
windows

[gitchat1]

Use either Brave or Vivaldi as your Browser and you won't have that problem. I don't think anyone ever really figured out what causes the issue.

[RBEmerson970]

@szaronko Please be more specific about browsers - just Chrome, a Chrome-based browser, or...? Also, which Windows version? And what hardware? FWIW, I regularly use Vivaldi to access Fast Koko at 127.0.0.1:8880 (under Docker, Win 11, i9 & RTX4090)

[szaronko]

@RBEmerson970
Windows 11. Ryzen 5600g, rtx3070, Tried brave, chrome firefox and edge. Somehow it worked just fine when I connected through edge with my laptop in the same local network, but it doesn't want to work on main pc. Now also tried vivaldi and still the same error

[RBEmerson970]

@szaronko If you're using Docker, how did you build the container?
I use
docker run --gpus all -p 127.0.0.1:8880:8880 ghcr.io/remsky/kokoro-fastapi-gpu:latest

Note I've added a specific "localhost" IP.

[szaronko]

@RBEmerson970
I used
docker run --gpus all -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-gpu:latest
as stated in the readme

[RBEmerson970]

@szaronko Nothing in the log looks wrong. NTL try building the container with the IP then connect to http://127.0.0.1:880/web

Unfortunately my AI machine is out for service (overheating problems - wonder why...) so I can't check the repo as it exists now.

Note version control on this repo sucx big time. You can only hope "latest" actually gets you the current mods and updates. If you're lucky, you should get v0.3.0.

[aindriu80]

I'm having the same problem with Firefox, I get the following message:

However I have been able to get voice to work in Firefox by using the Open Web UI interface which is linked to the same Kokoro instance, I can press on the audio button below an AI response and audio will start playing:-

It must be a issue with the Kokoro web interface

[gitchat1]

Did you use the prebuilt image or build it yourself.

[aindriu80]

I used the prebuilt command to get Kokoro up and running

docker run -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu

[gitchat1]

Try building it instead of using the prebuilt.

[aahmetblc906]

I am facing same issue with ghcr.io/remsky/kokoro-fastapi-cpu

[gitchat1]

Even after building it??

[wh0am1-dev]

I think the problem is with this line: AudioService.js#L135

if (this.mediaSource.readyState === 'open') {
  this.mediaSource.endOfStream();
}

It's calling this.mediaSource.endOfStream() with this.sourceBuffer.updating == true, which is not allowed.

Based on the example on MDN, I think it should call endOfStream() when the updateend event of the source buffer is triggered:

this.sourceBuffer.addEventListener("updateend", () => {
  this.mediaSource.endOfStream();
});

cc. @remsky

[MrTimonM]

i am getting this too

[kooshchic]

This web player does not work in Firefox due to Gemini 3.1 pro analysis : technical streaming restrictions (MediaSource API). Alternative as already mentioned use a Chromium-based browser Google Chrome.

Alternatively: Kokoro Reader (Firefox Add-on)
If you would like to use your local Kokoro server across the entire web rather than just for chats in the WebUI, there is an extremely convenient extension for Firefox: 👉 Kokoro Reader Add-on for Firefox

With this tool, you can highlight any passage of text or entire webpages and have them read aloud by your own local AI server.

Setup Magic: Installation is completely seamless. The add-on detects the environment and usually automatically fills in the local API address (http://127.0.0.1:8800/v1/) in the settings (or preferences/options). As a rule, you don't need to configure anything else and can get started immediately!

Have Fun