Update app.py

app.py

@@ -159,8 +159,10 @@ async def tts_say_stream_wav(
     write_event({"type":"tts_stream_get","len":len(text),"voice":voice,"ls":ls})
 
     async def gen():
-        ws_url = _tts_ws_url()
-        async with websockets.connect(ws_url, ping_interval=None, max_size=8_000_000) as ws:
+        ws = None
+        try:
+            ws_url = _tts_ws_url()
+            ws = await websockets.connect(ws_url, ping_interval=None, max_size=8_000_000)
             # init
             await ws.send(json.dumps({
                 "event": "init",
@@ -170,11 +172,11 @@ async def tts_say_stream_wav(
                 "noise_w": noise_w,
             }))
             sr, ch = 22050, 1
-            # wait for ready -> send header
+            # wait for ready -> send header immediately so client can start
             while True:
                 m = await ws.recv()
                 if isinstance(m, (bytes, bytearray)):
-                    # unlikely before ready; ignore
+                    # ignore stray audio until we know sr/ch
                     continue
                 try:
                     evt = json.loads(m)
@@ -196,20 +198,43 @@ async def tts_say_stream_wav(
             while True:
                 try:
                     msg = await ws.recv()
-                except websockets.exceptions.ConnectionClosed:
+                except websockets.exceptions.ConnectionClosedOK:
+                    break
+                except websockets.exceptions.ConnectionClosedError:
                     break
+
                 if isinstance(msg, (bytes, bytearray)):
-                    yield msg
-                else:
-                    try:
-                        evt = json.loads(msg)
-                    except Exception:
-                        continue
-                    if evt.get("event") in ("done", "end"):
-                        break
-                    if evt.get("event") == "error":
-                        yield f'ERROR: {evt.get("detail","tts error")}'.encode("utf-8")
-                        break
+                    # raw PCM16 frame from TTS; just yield
+                    if msg:
+                        yield msg
+                    continue
+
+                # control event
+                try:
+                    evt = json.loads(msg)
+                except Exception:
+                    continue
+                kind = evt.get("event")
+                if kind in ("done", "end"):
+                    break
+                if kind == "error":
+                    # propagate as bytes (don’t raise; avoid chunk abort)
+                    detail = evt.get("detail", "tts error")
+                    yield f'ERROR: {detail}'.encode("utf-8")
+                    break
+                # ignore logs
+
+        except Exception as e:
+            # Log, but don't raise (raising here aborts chunked stream & causes 'incomplete chunked read')
+            write_event({"type":"tts_stream_err","err":str(e)})
+            # a tiny trailing pad keeps some clients happy
+            yield b""
+        finally:
+            try:
+                if ws is not None:
+                    await ws.close()
+            except Exception:
+                pass
 
     return StreamingResponse(gen(), media_type="audio/wav",
                              headers={"Cache-Control":"no-cache","Connection":"keep-alive"})