open-webui/desktop

Fork 0

mirror of https://github.com/open-webui/desktop.git synced 2026-11-05 01:27:46 +00:00

[GH-ISSUE #98] enh: wakeword detection #1

New issue

Open

opened 2026-10-18 14:04:59 +00:00 by bsakaguchi · 5 comments

bsakaguchi commented

2026-10-18 14:04:59 +00:00

Owner

Originally created by @tjbck on GitHub (Jul 2, 2024).
Original GitHub issue: https://github.com/open-webui/desktop/issues/98

"Hey assistant" should toggle on the voice call.

Originally created by @tjbck on GitHub (Jul 2, 2024). Original GitHub issue: https://github.com/open-webui/desktop/issues/98 "Hey assistant" should toggle on the voice call.

bsakaguchi added the

help wanted

label

2026-10-18 14:04:59 +00:00

bsakaguchi referenced this issue

2026-10-18 14:04:59 +00:00

[GH-ISSUE #1] is this gonna be a react native mobile app for local models on smartphones? #3

bsakaguchi commented

2026-10-18 14:05:00 +00:00

Author

Owner

@moblangeois commented on GitHub (Jul 4, 2024):

Hello,

I came across a project that might be relevant to this issue regarding wake word detection. The project is called GLaDOS (https://github.com/dnhkng/GlaDOS).

While browsing their code, I found an interesting implementation of voice recognition and wake word detection that could potentially be adapted for our "Hey assistant" feature. You can find the relevant code here:

github.com/dnhkng/GlaDOS@99fbf5e182/glados/voice_recognition.py (L159)

Some key points about this implementation:

It uses a Voice Activity Detection (VAD) model to detect when speech is occurring.
It employs an Automatic Speech Recognition (ASR) model for transcription.
It includes a wake word detection method using Levenshtein distance to allow for slight variations in pronunciation.

The project is under the MIT license, which is permissive and should allow us to adapt parts of their code if we decide it's useful for our implementation.

I believe this could serve as a good starting point.

@moblangeois commented on GitHub (Jul 4, 2024): Hello, I came across a project that might be relevant to this issue regarding wake word detection. The project is called GLaDOS (https://github.com/dnhkng/GlaDOS). While browsing their code, I found an interesting implementation of voice recognition and wake word detection that could potentially be adapted for our "Hey assistant" feature. You can find the relevant code here: https://github.com/dnhkng/GlaDOS/blob/99fbf5e182ed2c60e075bd1241d67f68f6833b8e/glados/voice_recognition.py#L159 Some key points about this implementation: 1. It uses a Voice Activity Detection (VAD) model to detect when speech is occurring. 2. It employs an Automatic Speech Recognition (ASR) model for transcription. 3. It includes a wake word detection method using Levenshtein distance to allow for slight variations in pronunciation. The project is under the MIT license, which is permissive and should allow us to adapt parts of their code if we decide it's useful for our implementation. I believe this could serve as a good starting point.

bsakaguchi commented

2026-10-18 14:05:00 +00:00

Author

Owner

@jmtatsch commented on GitHub (Jul 5, 2024):

https://github.com/dscripka/openWakeWord

@jmtatsch commented on GitHub (Jul 5, 2024): https://github.com/dscripka/openWakeWord

bsakaguchi commented

2026-10-18 14:05:00 +00:00

Author

Owner

@moblangeois commented on GitHub (Jul 5, 2024):

https://github.com/dscripka/openWakeWord

Seems to use onnxruntime 'onnxruntime>=1.10.0,<2', while we use onnxruntime==1.17.3 for OCR it seems. Therefore it should be good.

@moblangeois commented on GitHub (Jul 5, 2024): > https://github.com/dscripka/openWakeWord Seems to use onnxruntime `'onnxruntime>=1.10.0,<2',` while we use `onnxruntime==1.17.3` for OCR it seems. Therefore it should be good.

bsakaguchi commented

2026-10-18 14:05:00 +00:00

Author

Owner

@tjbck commented on GitHub (Jul 5, 2024):

Requirements for this would be

ideally the model should run on the browser end
Should be easily customizable in the future to allow custom wakewords

I’ll add more to the list in the future.

@tjbck commented on GitHub (Jul 5, 2024): Requirements for this would be - ideally the model should run on the browser end - Should be easily customizable in the future to allow custom wakewords I’ll add more to the list in the future.

bsakaguchi commented

2026-10-18 14:05:00 +00:00

Author

Owner

@svdasein commented on GitHub (May 18, 2026):

Just curious - is the idea that the wake up word is recognized by something in the browser, or does the browser stream all the audio it receives to openwebui which then scans for the word? I do get that there's a desktop app in the works somewhere as well.

@svdasein commented on GitHub (May 18, 2026): Just curious - is the idea that the wake up word is recognized by something _in_ the browser, or does the browser stream all the audio it receives to openwebui which then scans for the word? I do get that there's a desktop app in the works somewhere as well.

bsakaguchi referenced this issue

2026-10-18 14:06:13 +00:00

[GH-ISSUE #178] Gray screen when connecting to remote server #110

bsakaguchi referenced this issue

2026-10-18 14:06:36 +00:00

[GH-ISSUE #249] Terminal connection shows as saved/connected in UI, but webui.db never actually stores it — and once you fix that by hand, chat tool-calls and the sidebar terminal disagree on what "the ID" even is #162

bsakaguchi referenced this issue

2026-10-18 14:07:13 +00:00

[PR #265] Add CUDA 13.3 and 13.4 variant support to Inference Runtime settings #236