auto-git:

[add] dist/assets/index-DifKT69O.js
 [change] backend/main.py
 [change] backend/schemas.py
 [change] backend/whisper_admin.py
 [change] dist/index.html
 [change] electron/main.cjs
 [change] src/App.jsx
 [change] src/GeneralSettings.jsx
 [change] src/audioInput.js
 [unlink] dist/assets/index-Dm7DZNSo.js

backend/main.py

@@ -159,6 +159,7 @@ async def transcribe_audio_route(req: schemas.AudioTranscriptionRequest):
             audio_bytes,
             mime_type,
             req.model or DEFAULT_WHISPER_MODEL,
+            req.language,
         )
     except RuntimeError as exc:
         raise HTTPException(status_code=400, detail=str(exc)) from exc

backend/schemas.py

@@ -84,6 +84,7 @@ class AudioTranscriptionRequest(BaseModel):
     mime_type: str
     audio_base64: str
     model: Optional[str] = None
+    language: Optional[str] = None
 
 
 class AudioTranscriptionResponse(BaseModel):

backend/whisper_admin.py

@@ -241,6 +241,7 @@ def transcribe_audio_bytes(
     audio_bytes: bytes,
     mime_type: str,
     model_name: str = DEFAULT_WHISPER_MODEL,
+    language: Optional[str] = None,
 ) -> Dict[str, Any]:
     if not audio_bytes:
         raise RuntimeError("Recorded audio was empty.")
@@ -256,7 +257,14 @@ def transcribe_audio_bytes(
             _convert_audio_to_wav(input_path, wav_path)
             source_path = wav_path
 
-        result = model.transcribe(str(source_path), task="transcribe", fp16=device == "cuda")
+        transcription_options = {
+            "task": "transcribe",
+            "fp16": device == "cuda",
+        }
+        if language:
+            transcription_options["language"] = str(language).strip().lower()
+
+        result = model.transcribe(str(source_path), **transcription_options)
         return {
             "model": model_name,
             "device": device,

dist/index.html

@@ -5,7 +5,7 @@
     <meta charset="UTF-8" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <title>LLM Desktop</title>
-    <script type="module" crossorigin src="/assets/index-Dm7DZNSo.js"></script>
+    <script type="module" crossorigin src="/assets/index-DifKT69O.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-BFOWWhCS.css">
   </head>
   <body>

electron/main.cjs

@@ -28,6 +28,7 @@ const MAX_UI_SCALE = 1.3
 const DEFAULT_OPEN_DEVTOOLS_ON_STARTUP = false
 const DEFAULT_AUDIO_INPUT_ENABLED = true
 const DEFAULT_AUDIO_INPUT_DEVICE_ID = ''
+const DEFAULT_AUDIO_INPUT_LANGUAGE = ''
 
 const defaultSettings = {
   backendApiUrl: DEFAULT_BACKEND_API_URL,
@@ -38,6 +39,7 @@ const defaultSettings = {
   openDevToolsOnStartup: DEFAULT_OPEN_DEVTOOLS_ON_STARTUP,
   audioInputEnabled: DEFAULT_AUDIO_INPUT_ENABLED,
   audioInputDeviceId: DEFAULT_AUDIO_INPUT_DEVICE_ID,
+  audioInputLanguage: DEFAULT_AUDIO_INPUT_LANGUAGE,
   chatModel: 'llama3',
 }
 
@@ -92,6 +94,7 @@ function migrateSettings(rawSettings) {
   nextSettings.openDevToolsOnStartup = normalizeOpenDevToolsOnStartup(nextSettings.openDevToolsOnStartup)
   nextSettings.audioInputEnabled = normalizeBooleanSetting(nextSettings.audioInputEnabled)
   nextSettings.audioInputDeviceId = String(nextSettings.audioInputDeviceId || '').trim()
+  nextSettings.audioInputLanguage = String(nextSettings.audioInputLanguage || '').trim().toLowerCase()
 
   return { nextSettings, migrated }
 }
@@ -553,6 +556,8 @@ ipcMain.handle('set-setting', (event, key, value) => {
     appSettings[key] = normalizeBooleanSetting(value)
   } else if (key === 'audioInputDeviceId') {
     appSettings[key] = String(value || '').trim()
+  } else if (key === 'audioInputLanguage') {
+    appSettings[key] = String(value || '').trim().toLowerCase()
   } else {
     appSettings[key] = value
   }
@@ -572,6 +577,7 @@ ipcMain.handle('update-settings', (event, settings) => {
   appSettings.openDevToolsOnStartup = normalizeOpenDevToolsOnStartup(appSettings.openDevToolsOnStartup)
   appSettings.audioInputEnabled = normalizeBooleanSetting(appSettings.audioInputEnabled)
   appSettings.audioInputDeviceId = String(appSettings.audioInputDeviceId || '').trim()
+  appSettings.audioInputLanguage = String(appSettings.audioInputLanguage || '').trim().toLowerCase()
   saveSettings()
   if (Object.prototype.hasOwnProperty.call(settings, 'uiScale')) {
     applyUiScaleToAllWindows()

src/App.jsx

@@ -295,6 +295,7 @@ export default function App() {
   const imageDragDepthRef = useRef(0)
   const [audioInputEnabled, setAudioInputEnabled] = useState(false)
   const [audioInputDeviceId, setAudioInputDeviceId] = useState('')
+  const [audioInputLanguage, setAudioInputLanguage] = useState('')
   const [isRecordingAudio, setIsRecordingAudio] = useState(false)
   const [isTranscribingAudio, setIsTranscribingAudio] = useState(false)
   const [audioRecordingMs, setAudioRecordingMs] = useState(0)
@@ -554,6 +555,7 @@ export default function App() {
         body: JSON.stringify({
           mime_type: mimeType || detectedMimeType || 'audio/webm',
           audio_base64: payload,
+          language: audioInputLanguage || null,
         }),
       })
       const data = await expectBackendJson(response)
@@ -1144,6 +1146,7 @@ async function regenerateFromIndex(index, overrideUserText = null) {
       setStreamOutput(settings.streamOutput || false);
       setAudioInputEnabled(settings.audioInputEnabled === true);
       setAudioInputDeviceId(typeof settings.audioInputDeviceId === 'string' ? settings.audioInputDeviceId : '');
+      setAudioInputLanguage(typeof settings.audioInputLanguage === 'string' ? settings.audioInputLanguage : '');
       setScrollPositions(settings.scrollPositions || {}); // Load scroll positions
       applyColorScheme(settings.colorScheme || 'Default'); // Apply initial scheme
     }).finally(() => {
@@ -2788,6 +2791,7 @@ async function createNewChat() {
                 onStreamOutputChange={setStreamOutput}
                 onAudioInputEnabledChange={setAudioInputEnabled}
                 onAudioInputDeviceChange={setAudioInputDeviceId}
+                onAudioInputLanguageChange={setAudioInputLanguage}
                 onLibrariesPurged={handleLibrariesPurged}
               />
             )}

src/GeneralSettings.jsx

@@ -2,6 +2,8 @@ import React, { useEffect, useState } from 'react';
 import {
   AUDIO_INPUT_DEVICE_ID_KEY,
   AUDIO_INPUT_ENABLED_KEY,
+  AUDIO_INPUT_LANGUAGE_KEY,
+  AUDIO_INPUT_LANGUAGE_OPTIONS,
   ensureAudioInputPermission,
   listAudioInputDevices,
   supportsAudioInputCapture,
@@ -13,6 +15,7 @@ const EMBED_MODEL_KEY = 'embedModel';
 const MODEL_KEY = 'chatModel';
 const STREAM_KEY = 'streamOutput';
 const DEFAULT_AUDIO_INPUT_DEVICE_ID = '';
+const DEFAULT_AUDIO_INPUT_LANGUAGE = '';
 const DEFAULT_BACKEND_API_URL = 'http://127.0.0.1:8000';
 const DEFAULT_OLLAMA_API_URL = 'http://127.0.0.1:11434';
 const DEFAULT_EMBED_MODEL = 'nomic-embed-text:latest';
@@ -47,6 +50,7 @@ export default function GeneralSettings({
   onBackendApiUrlChange,
   onAudioInputEnabledChange,
   onAudioInputDeviceChange,
+  onAudioInputLanguageChange,
 }) {
   const [backendApiUrl, setBackendApiUrl] = useState('');
   const [ollamaApiUrl, setOllamaApiUrl] = useState('');
@@ -56,6 +60,7 @@ export default function GeneralSettings({
   const [streamOutput, setStreamOutput] = useState(false);
   const [audioInputEnabled, setAudioInputEnabled] = useState(false);
   const [audioInputDeviceId, setAudioInputDeviceId] = useState(DEFAULT_AUDIO_INPUT_DEVICE_ID);
+  const [audioInputLanguage, setAudioInputLanguage] = useState(DEFAULT_AUDIO_INPUT_LANGUAGE);
   const [audioInputDevices, setAudioInputDevices] = useState([]);
   const [isRefreshingAudioDevices, setIsRefreshingAudioDevices] = useState(false);
   const [audioInputStatus, setAudioInputStatus] = useState({ tone: 'neutral', message: '' });
@@ -87,6 +92,11 @@ export default function GeneralSettings({
           ? settings.audioInputDeviceId
           : DEFAULT_AUDIO_INPUT_DEVICE_ID
       );
+      setAudioInputLanguage(
+        typeof settings.audioInputLanguage === 'string'
+          ? settings.audioInputLanguage
+          : DEFAULT_AUDIO_INPUT_LANGUAGE
+      );
       setUpdateStatus(status || DEFAULT_UPDATE_STATUS);
     });
 
@@ -268,6 +278,15 @@ export default function GeneralSettings({
     }
   };
 
+  const handleAudioInputLanguageChange = (event) => {
+    const nextLanguage = event.target.value;
+    setAudioInputLanguage(nextLanguage);
+    window.electronAPI.setSetting(AUDIO_INPUT_LANGUAGE_KEY, nextLanguage);
+    if (onAudioInputLanguageChange) {
+      onAudioInputLanguageChange(nextLanguage);
+    }
+  };
+
   const handleCheckForUpdates = async () => {
     setIsCheckingForUpdates(true);
     try {
@@ -425,6 +444,18 @@ export default function GeneralSettings({
                   </option>
                 ))}
               </select>
+              <select
+                className="select"
+                value={audioInputLanguage}
+                onChange={handleAudioInputLanguageChange}
+                disabled={!audioInputSupported}
+              >
+                {AUDIO_INPUT_LANGUAGE_OPTIONS.map(language => (
+                  <option key={language.value || 'auto'} value={language.value}>
+                    {language.label}
+                  </option>
+                ))}
+              </select>
               <button
                 type="button"
                 className="button"
@@ -437,6 +468,9 @@ export default function GeneralSettings({
             {audioInputStatus.message && (
               <p className={`setting-status ${audioInputStatus.tone}`}>{audioInputStatus.message}</p>
             )}
+            <p className="setting-description">
+              Whisper can auto-detect the spoken language, but you can force a fixed input language here when auto-detection drifts.
+            </p>
           </>
         )}
       </div>

src/audioInput.js

@@ -1,5 +1,22 @@
 export const AUDIO_INPUT_ENABLED_KEY = 'audioInputEnabled'
 export const AUDIO_INPUT_DEVICE_ID_KEY = 'audioInputDeviceId'
+export const AUDIO_INPUT_LANGUAGE_KEY = 'audioInputLanguage'
+
+export const AUDIO_INPUT_LANGUAGE_OPTIONS = [
+  { value: '', label: 'Auto' },
+  { value: 'de', label: 'German' },
+  { value: 'en', label: 'English' },
+  { value: 'fr', label: 'French' },
+  { value: 'es', label: 'Spanish' },
+  { value: 'it', label: 'Italian' },
+  { value: 'pt', label: 'Portuguese' },
+  { value: 'nl', label: 'Dutch' },
+  { value: 'pl', label: 'Polish' },
+  { value: 'tr', label: 'Turkish' },
+  { value: 'ru', label: 'Russian' },
+  { value: 'ja', label: 'Japanese' },
+  { value: 'zh', label: 'Chinese' },
+]
 
 const AUDIO_RECORDER_MIME_CANDIDATES = [
   'audio/webm;codecs=opus',