[GH-ISSUE #18104] feat: deepgram doesn't seem to work in French #18496

New Issue

GiteaMirror · 2026-04-20T00:43:38-05:00

GiteaMirror commented

2026-04-20 00:43:38 -05:00

Originally created by @scwall on GitHub (Oct 7, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/18104

Check Existing Issues

I have searched the existing issues and discussions.

Problem Description

Description

Thank you for this OpenWebUI project that I actively use 🙌
I would like to suggest an improvement regarding the speech recognition module (Speech-to-Text).

Currently, when using external engines like Deepgram, it's impossible to manually set the language or specify language=multi to enable multilingual recognition (for example, French + German in the same sentence).

Current Issue

The "Speech-to-Text Model" field does allow you to choose a model like nova-3,
but additional parameters (such as language=multi) cannot be set.
Some API calls such as those sent to Deepgram look like:
And it is not possible to add such correct calls:

Desired Solution you'd like

Add in the STT configuration interface:

✅ A free "Language" field, which would allow specifying the language or multi

Example for Deepgram: language=multi, language=fr, etc.

✅ Or a checkbox "Automatically detect language (multilingual)"
✅ Or a field "additional STT settings" to pass custom parameters to the call URL (/listen)

Use Case

Here is a typical example:

I want my voice assistant to understand phrases like:

"Bonjour, wie geht's dir heute?"

Without being able to set language=multi, the STT engine only recognizes English by default (Deepgram).

Thank's 🙏🙏🙏

Originally created by @scwall on GitHub (Oct 7, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/18104 ### Check Existing Issues - [x] I have searched the existing issues and discussions. ### Problem Description ## Description Thank you for this OpenWebUI project that I actively use 🙌 I would like to suggest an improvement regarding the speech recognition module (Speech-to-Text). Currently, when using external engines like Deepgram, it's impossible to manually set the language or specify `language=multi` to enable **multilingual** recognition (for example, French + German in the same sentence). ## Current Issue - The "Speech-to-Text Model" field does allow you to choose a model like `nova-3`, but **additional parameters (such as `language=multi`) cannot be set**. - Some API calls such as those sent to Deepgram look like: - And it is not possible to add such correct calls: ### Desired Solution you'd like Add in the STT configuration interface: 1. ✅ A free "Language" field, which would allow specifying the language or `multi` - Example for Deepgram: `language=multi`, `language=fr`, etc. 2. ✅ Or a checkbox "Automatically detect language (multilingual)" 3. ✅ Or a field "additional STT settings" to pass custom parameters to the call URL (/listen) ## Use Case Here is a typical example: - I want my voice assistant to understand phrases like: > *"Bonjour, wie geht's dir heute?"* - Without being able to set `language=multi`, **the STT engine only recognizes English by default** (Deepgram). ## Thank's 🙏🙏🙏

GiteaMirror closed this issue

2026-04-20 00:43:38 -05:00

GiteaMirror referenced this issue

2026-04-20 05:36:11 -05:00

[PR #18496] [MERGED] fix: Clear file list when selected models do not support file uploads #24815

GiteaMirror referenced this issue

2026-04-25 12:55:42 -05:00

[PR #18496] [MERGED] fix: Clear file list when selected models do not support file uploads #40445

GiteaMirror referenced this issue

2026-04-29 23:16:01 -05:00

[PR #18496] [MERGED] fix: Clear file list when selected models do not support file uploads #47863

GiteaMirror referenced this issue