improved rvc api

Author: SayanoAI

Home.py

@@ -21,10 +21,11 @@
 
 from webui.contexts import ProgressBarContext, SessionStateContext
 
-def download_audio_to_buffer(url):
+@st.cache_data
+def download_audio_to_buffer(url,subtype="mp4"):
     buffer = BytesIO()
     youtube_video = YouTube(url)
-    audio = youtube_video.streams.get_audio_only()
+    audio = youtube_video.streams.get_audio_only(subtype=subtype)
     default_filename = slugify_filepath(audio.default_filename)
     audio.stream_to_buffer(buffer)
     return default_filename, buffer

api.py

@@ -24,11 +24,11 @@ async def get_rvc():
     gc_collect()
     return list_rvc_models()
 
-@server.post("/rvc")
-async def rvc_infer(body: RVCInferenceParams):
+@server.post("/rvc/{name}")
+async def rvc_infer(body: RVCInferenceParams, name: str):
     response = {}
     gc_collect()
-    output_audio = convert_vocals(**vars(body))
+    output_audio = convert_vocals(name=name,**vars(body))
     if output_audio: response["data"] = audio2bytes(*output_audio)
     gc_collect()
     return response

lib/audio.py

@@ -99,24 +99,6 @@ def bytes_to_audio(data: Union[io.BytesIO,bytes],**kwargs):
             audio = audio.T # transpose to channels-first
     return audio, sr
 
-# def bytes2audio(data: str):
-#     try:
-#         iofile = io.BytesIO(base64.b64decode(data))
-#         decoded = np.load(iofile)
-#         return decoded["audio"], decoded["sr"]+0
-#     except Exception as e:
-#         print(e)
-#     return None
-
-# def audio2bytes(audio: np.array, sr: int):
-#     try:
-#         iofile = io.BytesIO()
-#         np.savez_compressed(iofile,audio=audio,sr=sr)
-#         return base64.b64encode(iofile.getvalue()).decode("utf-8")
-#     except Exception as e:
-#         print(e)
-#     return ""
-
 def bytes2audio(data: str):
     try:
         # Split the suffixed data by the colon

lib/utils.py

@@ -68,7 +68,7 @@ def get_optimal_torch_device(index = 0) -> torch.device:
 
 def get_optimal_threads(offset=0):
     cores = multiprocessing.cpu_count() - offset
-    return max(np.floor(cores * (1-psutil.cpu_percent())),1)
+    return int(max(np.floor(cores * (1-psutil.cpu_percent())),1))
 
 def pid_is_active(pid: int):        
     """ Check For the existence of a unix pid. https://stackoverflow.com/a/568285"""

server/types.py

@@ -3,7 +3,6 @@
 
 
 class RVCInferenceParams(BaseModel):
-    name: str
     audio_data: str
     f0_up_key: int=0
     f0_method: List[Literal["rmvpe","rmvpe+","crepe","mangio-crepe"]]=["rmvpe"]

webui/api.py

@@ -67,9 +67,9 @@ def split_vocals(audio_path,**args):
 def convert_vocals(model_name,input_audio,**kwargs):
     gc_collect()
     audio_data = audio2bytes(*input_audio)
-    body = dict(name=model_name,audio_data=audio_data,**kwargs)
+    body = dict(audio_data=audio_data,**kwargs)
 
-    with requests.post(RVC_INFERENCE_URL,json=body) as req:
+    with requests.post(f"{RVC_INFERENCE_URL}/{model_name}",json=body) as req:
         if req.status_code==200:
             response = req.json()
             audio = bytes2audio(response["data"])