fix: 修复牛牛唱歌失效问题

app/api/endpoints/sing.py

@@ -12,7 +12,7 @@ async def sing_endpoint(request_id: str, request: SingRequest):
     return SingResponse(task_id=task_id, status="processing")
 
 
-@router.get("/play/{request_id}", response_model=SingResponse)
-async def play_endpoint(request_id: str, speaker: str = ""):
-    await play(request_id, speaker)
-    return SingResponse(task_id="", status="processing")
+@router.get("/play/{speaker}", response_model=SingResponse)
+async def play_endpoint(speaker: str):
+    task_id = await play(speaker)
+    return SingResponse(task_id=task_id, status="processing")

app/services/callback.py

@@ -31,3 +31,17 @@ async def callback_audio(request_id: str, audio: bytes):
         {"status": "success"},
         files={"file": audio},
     )
+
+
+async def callback_audio_with_info(request_id: str, audio: bytes, song_id: str, chunk_index: int, key: int):
+    callback_url = f"{CALLBACK_URL}/{request_id}"
+    await send_callback(
+        callback_url,
+        {
+            "status": "success",
+            "song_id": song_id,
+            "chunk_index": chunk_index,
+            "key": key,
+        },
+        files={"file": audio},
+    )

app/services/sing.py

@@ -1,13 +1,8 @@
-import random
-from pathlib import Path
+from ulid import ULID
 
 from app.core.config import settings
 from app.core.logger import logger
-from app.services.callback import callback_audio, callback_failed
-from app.tasks.sing import sing_task
-
-SONG_PATH = "resource/sing/splices/"
-MUSIC_PATH = "resource/music/"
+from app.tasks.sing import play_task, sing_task
 
 
 async def sing(request_id: str, speaker: str, song_id: int, key: int, chunk_index: int):
@@ -16,38 +11,8 @@ async def sing(request_id: str, speaker: str, song_id: int, key: int, chunk_inde
     return task.id
 
 
-def get_random_song(speaker: str = ""):
-    all_song = []
-    if Path(SONG_PATH).exists():
-        all_song = [
-            str(s)
-            for s in Path(SONG_PATH).iterdir()
-            # 只唱过一段的大概率不是什么好听的，排除下
-            if speaker in s.name and "_spliced0" not in s.name
-        ]
-    if not all_song:
-        all_song = [MUSIC_PATH + s for s in Path(MUSIC_PATH).iterdir()]
-
-    if not all_song:
-        return None
-    return random.choice(all_song)
-
-
 async def play(speaker: str = ""):
-    rand_music = get_random_song(speaker)
-    if not rand_music:
-        await callback_failed()
-        return
-
-    if "_spliced" in rand_music:
-        splited = Path(rand_music).stem.split("_")
-        song_id = splited[0]
-        chunk_index = int(splited[1].replace("spliced", "")) + 1
-    elif "_full_" in rand_music:
-        song_id = Path(rand_music).stem.split("_")[0]
-        chunk_index = 114514
-    else:
-        song_id = ""
-        chunk_index = 114514
-
-    await callback_audio(speaker, song_id, 0, chunk_index, rand_music)
+    request_id = str(ULID())
+    task = play_task.delay(request_id, speaker)
+    logger.info(f"Task {task.id} started")
+    return request_id

app/tasks/__init__.py

@@ -1,2 +1,3 @@
 from app.tasks.chat.chat_tasks import chat_task
+from app.tasks.sing.play_tasks import play_task
 from app.tasks.sing.sing_tasks import sing_task

app/tasks/sing/__init__.py

@@ -1 +1,2 @@
+from app.tasks.sing.play_tasks import play_task
 from app.tasks.sing.sing_tasks import sing_task

app/tasks/sing/play_tasks.py

@@ -0,0 +1,82 @@
+import asyncio
+import random
+from pathlib import Path
+
+import anyio
+
+from app.core.celery import celery_app
+from app.services.callback import callback_audio_with_info, callback_failed
+
+SONG_PATH = "resource/sing/splices/"
+MUSIC_PATH = "resource/music/"
+
+
+def get_random_song(speaker: str = ""):
+    all_song = []
+    song_dir = Path(SONG_PATH)
+    if song_dir.exists():
+        all_song = [str(s) for s in song_dir.iterdir() if speaker in s.name and "_spliced0" not in s.name]
+
+    if not all_song:
+        music_dir = Path(MUSIC_PATH)
+        if music_dir.exists():
+            all_song = [str(s) for s in music_dir.iterdir()]
+
+    if not all_song:
+        return None
+    return random.choice(all_song)
+
+
+@celery_app.task(name="play")
+def play_task(request_id: str, speaker: str = ""):
+    loop = asyncio.new_event_loop()
+    asyncio.set_event_loop(loop)
+    try:
+        return loop.run_until_complete(_play_task_async(request_id, speaker))
+    finally:
+        loop.close()
+
+
+async def _play_task_async(request_id: str, speaker: str = ""):
+    try:
+        rand_music = get_random_song(speaker)
+        if not rand_music:
+            await callback_failed(request_id)
+            return False
+
+        if "_spliced" in rand_music:
+            splited = Path(rand_music).stem.split("_")
+            song_id = splited[0]
+            chunk_index = int(splited[1].replace("spliced", "")) + 1
+            if "key" in rand_music:
+                key_index = next((i for i, part in enumerate(splited) if "key" in part), None)
+                if key_index is not None:
+                    try:
+                        key = int(splited[key_index].replace("key", ""))
+                    except ValueError:
+                        key = 0
+                else:
+                    key = 0
+            else:
+                key = 0
+        elif "_full_" in rand_music:
+            song_id = Path(rand_music).stem.split("_")[0]
+            chunk_index = 114514
+            key = 0
+        else:
+            song_id = ""
+            chunk_index = 114514
+            key = 0
+
+        try:
+            async with await anyio.open_file(rand_music, "rb") as f:
+                audio_content = await f.read()
+        except Exception:
+            await callback_failed(request_id)
+            return False
+
+        await callback_audio_with_info(request_id, audio_content, song_id=song_id, chunk_index=chunk_index, key=key)
+        return True
+    except Exception:
+        await callback_failed(request_id)
+        return False

pyproject.toml

@@ -49,6 +49,7 @@ sing = [
     "torchfcpe>=0.0.4",
     "tqdm>=4.66.5",
     "transformers>=4.49.0",
+    "python-ulid>=3.0.0",
 ]
 tts = [
     "numpy==1.26.4",
@@ -70,6 +71,7 @@ tts = [
     "cn2an>=0.5.23",
     "pyopenjtalk>=0.4.1",
     "opencc>=1.1.9",
+    "tensorboardx==2.6.4",
 ]
 
 [tool.uv]