From f158f07ee0239c2e60f3f393e8d1447f6156810d Mon Sep 17 00:00:00 2001
From: Mieszko Makuch <m.makuch@cyfronet.pl>
Date: Wed, 25 Mar 2026 11:38:57 +0100
Subject: [PATCH] feat(stt): add server-side language fallback and extraParams
 for OpenAI STT provider

Add two optional fields to the OpenAI STT provider config schema:

- `language`: server-side default language (ISO 639-1) sent to Whisper when
  the client doesn't provide one. Useful for non-English deployments where
  admins want to predefine the transcription language without requiring each
  user to configure it in the browser.

- `extraParams`: arbitrary key-value pairs forwarded to the STT endpoint.
  Enables self-hosted Whisper servers (e.g. Speaches, faster-whisper-server)
  to receive provider-specific parameters like `vad_filter` (Voice Activity
  Detection) which filters silence and prevents hallucinations on empty
  audio clips. These params are ignored by the official OpenAI API.

Example librechat.yaml configuration:

```yaml
speech:
  stt:
    openai:
      url: 'http://whisper-server/v1/audio/transcriptions'
      apiKey: 'none'
      model: 'whisper-large-v3-turbo'
      language: 'pl'
      extraParams:
        vad_filter: true
```
---
 api/server/services/Files/Audio/STTService.js | 6 +++++-
 packages/data-provider/src/config.ts          | 2 ++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/api/server/services/Files/Audio/STTService.js b/api/server/services/Files/Audio/STTService.js
index 4ba62a7eeb..3a7a51bd8d 100644
--- a/api/server/services/Files/Audio/STTService.js
+++ b/api/server/services/Files/Audio/STTService.js
@@ -206,6 +206,10 @@ class STTService {
       data.language = validLanguage;
     }
 
+    if (sttSchema?.extraParams) {
+      Object.assign(data, sttSchema.extraParams);
+    }
+
     const headers = {
       'Content-Type': 'multipart/form-data',
       ...(apiKey && { Authorization: `Bearer ${apiKey}` }),
@@ -338,7 +342,7 @@ class STTService {
 
     try {
       const [provider, sttSchema] = await this.getProviderSchema(req);
-      const language = req.body?.language || '';
+      const language = req.body?.language || sttSchema?.language || '';
       const text = await this.sttRequest(provider, sttSchema, { audioBuffer, audioFile, language });
       res.json({ text });
     } catch (error) {
diff --git a/packages/data-provider/src/config.ts b/packages/data-provider/src/config.ts
index 9bc3822c4b..fb89c7c486 100644
--- a/packages/data-provider/src/config.ts
+++ b/packages/data-provider/src/config.ts
@@ -478,6 +478,8 @@ const sttOpenaiSchema = z.object({
   url: z.string().optional(),
   apiKey: z.string(),
   model: z.string(),
+  language: z.string().optional(),
+  extraParams: z.record(z.unknown()).optional(),
 });
 
 const sttAzureOpenAISchema = z.object({