exception handling

melo/api.py

@@ -74,10 +74,14 @@ def audio_numpy_concat(segment_data_list, sr, speed=1.):
     @staticmethod
     def split_sentences_into_pieces(text, language, quiet=False):
         texts = split_sentence(text, language_str=language)
+        if len(texts) == 0:
+            return ""
+
         if not quiet:
             print(" > Text split to sentences.")
             print('\n'.join(texts))
             print(" > ===========================")
+        print(text)
         return texts
 
     def tts_to_file(self, text, speaker_id, output_path=None, sdp_ratio=0.2, noise_scale=0.6, noise_scale_w=0.8, speed=1.0, pbar=None, format=None, position=None, quiet=False,):
@@ -138,6 +142,8 @@ def synthesize(self, text, speaker_id, output_path=None, sdp_ratio=0.2, noise_sc
         texts = self.split_sentences_into_pieces(text, language, quiet)
         audio_list = []
         for t in texts:
+            if t is None:
+                continue
             if language in ['EN', 'ZH_MIX_EN']:
                 t = re.sub(r'([a-z])([A-Z])', r'\1 \2', t)
             device = self.device

server.py

@@ -0,0 +1,63 @@
+from melo.api import TTS
+from fastapi import FastAPI
+import uvicorn
+import numpy as np
+from pydantic import BaseModel
+from base64 import b64encode, b64decode
+from contextlib import asynccontextmanager
+import torch
+import torchaudio.transforms as T
+from typing import Optional
+import time
+from utils import write_bytesIO
+
+TTS_Server = None
+speaker_ids = None
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    global TTS_Server
+    global speaker_ids
+    # Load the ML model
+    TTS_Server = TTS('EN')
+    speaker_ids = TTS_Server.hps.data.spk2id
+    yield
+    # Clean up the ML models and release the resources
+    TTS_Server.close()
+
+class TTSResponse(BaseModel):
+    voice_id:str
+    text:str
+    sr:int
+    sdp_ratio:Optional[float] = 0.2
+    noise_scale:Optional[float] = 0.6
+    noise_scale_w:Optional[float] =  0.8
+    speed:Optional[float] = 1.0
+
+app = FastAPI(lifespan=lifespan)
+
+
+
+@app.post("/connection")
+def tts_process(response:TTSResponse):
+    __t = time.time()
+    try:
+        audio, sr = TTS_Server.synthesize(response.text,speaker_id=speaker_ids[response.voice_id], sdp_ratio=response.sdp_ratio, noise_scale=response.noise_scale, noise_scale_w=response.noise_scale_w, speed=response.speed)
+        audio = torch.from_numpy(audio)
+        resampler = T.Resample(sr, response.sr, dtype=audio.dtype)
+        audio = resampler(audio)
+        audio = audio.detach().numpy()
+        files = write_bytesIO(response.sr,audio)
+        return {'audio': b64encode(files.read()).decode(),'sr':response.sr,"time":time.time() - __t}
+    except Exception as e:
+        print(f"Something went wrong {e}")
+        with open('empty.wav', 'rb') as f:
+            audio = f.read()
+        resampler = T.Resample(44100, response.sr, dtype=np.int16)
+        audio = audio.detach().numpy()
+        files = write_bytesIO(response.sr,audio)
+        return {'audio': b64encode(files.read()).decode(),'sr':response.sr,"time":time.time() - __t}
+
+
+
+if __name__ == "__main__":
+    uvicorn.run("server:app",host='0.0.0.0',port=8000,reload=True)