diff --git a/src/servers/api/stt/transcribe.ts b/src/servers/api/stt/transcribe.ts
index 841c8684..a93e187e 100644
--- a/src/servers/api/stt/transcribe.ts
+++ b/src/servers/api/stt/transcribe.ts
@@ -4,6 +4,10 @@
 // Only ask Whisper to translate (which is English-only) when the detected
 // language is NOT one the user speaks — i.e. they wouldn't understand the
 // raw transcript anyway.
+//
+// The detection must win when Whisper is confident. Biasing toward the user's spoken languages
+// unconditionally means a user who speaks only English forces `language=en` onto every recording,
+// and Whisper answers non-English audio with a rough, unmarked translation instead of a transcript.
 
 type TranscribeArgs = {
   file: File | Blob;
@@ -17,6 +21,10 @@ export type TranscribeResult = {
   translated: boolean;
 };
 
+// Above this probability Whisper's own detection is taken as authoritative. Below it the clip is
+// treated as ambiguous and the user's spoken languages break the tie.
+const CONFIDENT_DETECTION = 0.5;
+
 // whisper.cpp's "language" field uses full names ("english"); ISO codes live in
 // `language_probabilities`. Map covers what the user can pick in Profile settings.
 const LANG_NAME_TO_ISO: Record<string, string> = {
@@ -33,11 +41,14 @@ const LANG_NAME_TO_ISO: Record<string, string> = {
   korean: 'ko',
 };
 
-export async function transcribeAudio({ file, whisperUrl, spokenLanguages }: TranscribeArgs): Promise<TranscribeResult> {
+export async function transcribeAudio({
+  file,
+  whisperUrl,
+  spokenLanguages,
+}: TranscribeArgs): Promise<TranscribeResult> {
   const base = whisperUrl.replace(/\/+$/, '');
 
-  // Step 1: detect language. Bias toward the user's spoken languages — Whisper's
-  // raw auto-detect is unreliable on short clips and often picks English by default.
+  // Step 1: detect language, and only fall back to the spoken-language bias when Whisper is unsure.
   const detectForm = new FormData();
   detectForm.append('file', file);
   detectForm.append('temperature', '0.0');
@@ -48,24 +59,32 @@ export async function transcribeAudio({ file, whisperUrl, spokenLanguages }: Tra
   if (!detectRes.ok) throw new Error(`Language detection failed (${detectRes.status})`);
   const detectJson = (await detectRes.json()) as {
     language?: string;
+    detected_language?: string;
+    detected_language_probability?: number;
     language_probabilities?: Record<string, number>;
   };
 
   const probs = detectJson.language_probabilities;
+  const ranked = probs ? Object.entries(probs).sort((a, b) => b[1] - a[1]) : [];
+  const [topCode, topProb] = ranked[0] ?? [];
+
   let detectedLanguage: string;
-  if (probs && spokenLanguages.length > 0) {
-    // Pick the spoken language with the highest probability
-    const ranked = spokenLanguages
-      .map((code) => [code, probs[code] ?? 0] as const)
-      .sort((a, b) => b[1] - a[1]);
-    detectedLanguage = ranked[0]?.[0] ?? spokenLanguages[0]!;
-  } else if (probs) {
-    // No spoken-language hint: take the highest-probability ISO code
-    const top = Object.entries(probs).sort((a, b) => b[1] - a[1])[0];
-    detectedLanguage = top?.[0] ?? 'en';
+  if (topCode && (topProb ?? 0) >= CONFIDENT_DETECTION) {
+    // Whisper is sure. Trust it even when it is not a language the user speaks — that is exactly the
+    // case translation exists for.
+    detectedLanguage = topCode;
+  } else if (ranked.length > 0 && spokenLanguages.length > 0) {
+    // Genuinely ambiguous (usually a short clip). Break the tie toward a language the user speaks,
+    // since Whisper's low-confidence guess defaults to English regardless of the audio.
+    const bestSpoken = spokenLanguages
+      .map((code) => [code, probs?.[code] ?? 0] as const)
+      .sort((a, b) => b[1] - a[1])[0];
+    detectedLanguage = bestSpoken?.[0] ?? topCode ?? 'en';
+  } else if (topCode) {
+    detectedLanguage = topCode;
   } else {
-    // No probabilities field (non-whisper.cpp backend) — try to map the name
-    const raw = (detectJson.language ?? 'en').toLowerCase();
+    // No probabilities field (non-whisper.cpp backend) — try to map the reported name.
+    const raw = (detectJson.detected_language ?? detectJson.language ?? 'en').toLowerCase();
     detectedLanguage = LANG_NAME_TO_ISO[raw] ?? raw;
   }