feat(ai/stt): add dictation language selection to STT settings

Add a per-workspace `sttLanguage` setting (ISO-639-1 hint; empty =
auto-detect) and a searchable language picker in the Voice / STT settings
card. The hint is forwarded to the transcription endpoint:
- multipart path via the AI SDK `providerOptions.openai.language`
- JSON (OpenRouter) path via a top-level `language` body field
only when non-empty, so auto-detect behaves exactly as before.

Threaded through the whole stack: ai.types, update DTO, AiSettingsService
(resolve/getMasked/update), the workspace.repo SQL allowlist, the client
ai-settings service types, and the provider-settings form. Adds en-US
source keys and ru-RU translations.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

apps/server/src/integrations/ai/ai-settings.service.ts

@@ -33,6 +33,8 @@ export interface UpdateAiSettingsInput {
   sttModel?: string;
   sttBaseUrl?: string;
   sttApiStyle?: SttApiStyle;
+  // ISO-639-1 dictation language hint (e.g. 'en', 'ru'). Empty = auto-detect.
+  sttLanguage?: string;
   sttApiKey?: string;
   publicShareChatModel?: string;
   publicShareAssistantRoleId?: string;
@@ -166,6 +168,8 @@ export class AiSettingsService {
       // Plain passthrough, no fallback; the transcribe path defaults unset to
       // 'multipart' (current behavior).
       sttApiStyle: provider.sttApiStyle,
+      // Plain passthrough; empty/unset = auto-detect at the transcribe path.
+      sttLanguage: provider.sttLanguage,
       baseUrl: provider.baseUrl,
       systemPrompt: provider.systemPrompt,
     };
@@ -240,6 +244,7 @@ export class AiSettingsService {
       sttModel: provider.sttModel,
       sttBaseUrl: provider.sttBaseUrl,
       sttApiStyle: provider.sttApiStyle,
+      sttLanguage: provider.sttLanguage,
       systemPrompt: provider.systemPrompt,
       publicShareChatModel: provider.publicShareChatModel,
       publicShareAssistantRoleId: provider.publicShareAssistantRoleId,
@@ -279,6 +284,7 @@ export class AiSettingsService {
       'sttModel',
       'sttBaseUrl',
       'sttApiStyle',
+      'sttLanguage',
       'systemPrompt',
       'publicShareChatModel',
       'publicShareAssistantRoleId',

apps/server/src/integrations/ai/ai.service.ts

@@ -212,12 +212,22 @@ export class AiService {
     const cfg = await this.aiSettings.resolve(workspaceId);
     if (!cfg?.sttModel) throw new AiSttNotConfiguredException();
     const baseURL = cfg.sttBaseUrl || cfg.baseUrl;
+    // Trimmed language hint; empty/unset = auto-detect (never forward an empty
+    // string to the provider, which would override auto-detect).
+    const sttLanguage = cfg.sttLanguage?.trim() || undefined;
 
     // Explicit, admin-chosen request encoding (no URL guessing). 'json' is the
     // OpenRouter style (JSON + base64 input_audio); everything else uses the
     // OpenAI-compatible multipart path via the AI SDK.
     if (cfg.sttApiStyle === 'json') {
-      return this.transcribeJsonBase64(baseURL, cfg.sttApiKey, cfg.sttModel, audio, format);
+      return this.transcribeJsonBase64(
+        baseURL,
+        cfg.sttApiKey,
+        cfg.sttModel,
+        audio,
+        format,
+        sttLanguage,
+      );
     }
 
     // Standard OpenAI-compatible multipart path (AI SDK). apiKey may be unused for
@@ -226,14 +236,23 @@ export class AiService {
       apiKey: cfg.sttApiKey ?? 'unused',
       baseURL,
     }).transcription(cfg.sttModel);
-    const { text } = await transcribe({ model, audio });
+    const { text } = await transcribe({
+      model,
+      audio,
+      // Forward the language hint only when set; the OpenAI transcription model
+      // reads it from providerOptions.openai.language.
+      ...(sttLanguage
+        ? { providerOptions: { openai: { language: sttLanguage } } }
+        : {}),
+    });
     return text.trim();
   }
 
   /**
    * JSON + base64 transcription body (OpenRouter-style). POSTs
    * { model, input_audio: { data, format } } to {baseURL}/audio/transcriptions
-   * and returns { text }.
+   * and returns { text }. The optional `language` ISO-639-1 hint is included as
+   * a top-level body field only when set (empty/unset = auto-detect).
    */
   private async transcribeJsonBase64(
     baseURL: string | undefined,
@@ -241,6 +260,7 @@ export class AiService {
     model: string,
     audio: Uint8Array,
     format: string,
+    language?: string,
   ): Promise<string> {
     if (!baseURL) {
       throw new BadRequestException(
@@ -256,6 +276,7 @@ export class AiService {
       },
       body: JSON.stringify({
         model,
+        ...(language ? { language } : {}),
         input_audio: {
           data: Buffer.from(audio).toString('base64'),
           format,

apps/server/src/integrations/ai/ai.types.ts

@@ -31,6 +31,8 @@ export interface AiProviderSettings {
   // STT-specific base URL. Falls back to baseUrl when empty/unset.
   sttBaseUrl?: string;
   sttApiStyle?: SttApiStyle;
+  // ISO-639-1 dictation language hint (e.g. 'en', 'ru'). Empty/unset = auto-detect.
+  sttLanguage?: string;
   systemPrompt?: string;
   // Cheap chat model id used ONLY by the anonymous public-share assistant. The
   // driver / baseUrl / apiKey of the main chat provider are reused; this is the
@@ -80,6 +82,8 @@ export interface MaskedAiSettings {
   sttModel?: string;
   sttBaseUrl?: string;
   sttApiStyle?: SttApiStyle;
+  // ISO-639-1 dictation language hint (e.g. 'en', 'ru'). Empty/unset = auto-detect.
+  sttLanguage?: string;
   systemPrompt?: string;
   publicShareChatModel?: string;
   // Agent-role id whose persona the public-share assistant adopts; empty/unset

apps/server/src/integrations/ai/dto/update-ai-settings.dto.ts

@@ -54,6 +54,11 @@ export class UpdateAiSettingsDto {
   @IsIn(STT_API_STYLES)
   sttApiStyle?: SttApiStyle;
 
+  // ISO-639-1 dictation language hint (e.g. 'en', 'ru'). Empty = auto-detect.
+  @IsOptional()
+  @IsString()
+  sttLanguage?: string;
+
   @IsOptional()
   @IsString()
   sttApiKey?: string;