Modification of the STT structure

apps/stt/main.py

@@ -177,27 +177,38 @@ def _wav_to_float32(wav_bytes: bytes) -> tuple[np.ndarray, int]:
     return audio, rate
 
 
+def _resample_audio(audio: np.ndarray, src_rate: int, dst_rate: int = 16000) -> np.ndarray:
+    if src_rate == dst_rate or audio.size == 0:
+        return audio.astype(np.float32, copy=False)
+    duration = audio.shape[0] / float(src_rate)
+    new_len = max(1, int(round(duration * dst_rate)))
+    x_old = np.linspace(0.0, 1.0, num=audio.shape[0], endpoint=False)
+    x_new = np.linspace(0.0, 1.0, num=new_len, endpoint=False)
+    return np.interp(x_new, x_old, audio).astype(np.float32)
+
+
 def _ensure_16k(wav_bytes: bytes) -> np.ndarray:
     audio, rate = _wav_to_float32(wav_bytes)
     if rate == 16000:
         return audio
 
+    # Prefer ffmpeg when available; otherwise pure-numpy resample (browser often sends 44.1/48 kHz)
     ffmpeg = shutil.which("ffmpeg")
-    if not ffmpeg:
-        raise HTTPException(status_code=400, detail="نرخ نمونه‌برداری صوت پشتیبانی نشد")
-    with tempfile.TemporaryDirectory() as tmp:
-        src = Path(tmp) / "in.wav"
-        dst = Path(tmp) / "out.wav"
-        src.write_bytes(wav_bytes)
-        result = subprocess.run(
-            [ffmpeg, "-y", "-i", str(src), "-ac", "1", "-ar", "16000", "-f", "wav", str(dst)],
-            capture_output=True,
-            check=False,
-        )
-        if result.returncode != 0 or not dst.exists():
-            raise HTTPException(status_code=400, detail="نرمال‌سازی صوت ناموفق بود")
-        audio, _ = _wav_to_float32(dst.read_bytes())
-        return audio
+    if ffmpeg:
+        with tempfile.TemporaryDirectory() as tmp:
+            src = Path(tmp) / "in.wav"
+            dst = Path(tmp) / "out.wav"
+            src.write_bytes(wav_bytes)
+            result = subprocess.run(
+                [ffmpeg, "-y", "-i", str(src), "-ac", "1", "-ar", "16000", "-f", "wav", str(dst)],
+                capture_output=True,
+                check=False,
+            )
+            if result.returncode == 0 and dst.exists():
+                audio, _ = _wav_to_float32(dst.read_bytes())
+                return audio
+
+    return _resample_audio(audio, rate, 16000)
 
 
 def _transcribe_local(wav_bytes: bytes) -> str:

apps/web/src/components/VoiceToTextButton.vue

@@ -137,6 +137,25 @@ function mergeChunks(parts: Float32Array[]) {
   return out;
 }
 
+/** Downsample to 16 kHz so STT does not depend on server ffmpeg. */
+function resampleTo16k(samples: Float32Array, srcRate: number) {
+  const dstRate = 16000;
+  if (!srcRate || srcRate === dstRate) return { samples, rate: srcRate || dstRate };
+  const ratio = srcRate / dstRate;
+  const newLen = Math.max(1, Math.floor(samples.length / ratio));
+  const out = new Float32Array(newLen);
+  for (let i = 0; i < newLen; i += 1) {
+    const srcIndex = i * ratio;
+    const i0 = Math.floor(srcIndex);
+    const i1 = Math.min(samples.length - 1, i0 + 1);
+    const t = srcIndex - i0;
+    const a = samples[i0] ?? 0;
+    const b = samples[i1] ?? a;
+    out[i] = a + (b - a) * t;
+  }
+  return { samples: out, rate: dstRate };
+}
+
 function encodeWav(samples: Float32Array, rate: number) {
   const bytes = samples.length * 2;
   const buffer = new ArrayBuffer(44 + bytes);
@@ -189,7 +208,8 @@ async function finish(keep: boolean) {
 
   setTranscribing(true);
   try {
-    const text = await upload(encodeWav(samples, rate));
+    const prepared = resampleTo16k(samples, rate);
+    const text = await upload(encodeWav(prepared.samples, prepared.rate));
     if (!text) {
       setError('متنی از صدا تشخیص داده نشد.');
       return;