Fix: endpoint passing speaker.

handler.py

@@ -7,9 +7,9 @@ from transformers import AutoModel, AutoProcessor
 class EndpointHandler:
     def __init__(self, path=""):
         # load model and processor from path
-        self.processor = AutoProcessor.from_pretrained(path)
+        self.processor = AutoProcessor.from_pretrained("suno/bark")
         self.model = AutoModel.from_pretrained(
-            path,
+            "suno/bark",
         ).to("cuda")
 
     def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
@@ -20,14 +20,18 @@ class EndpointHandler:
         """
         # process input
         text = data.pop("inputs", data)
-        parameters = data.get("parameters", None)
-
-        # preprocess
-        inputs = self.processor(
-            text=[text],
-            return_tensors="pt",
-            voice_preset=parameters.get("voice_preset", None),
-        ).to("cuda")
+        voice_preset = data.get("voice_preset", None)
+        if voice_preset:
+            inputs = self.processor(
+                text=[text],
+                return_tensors="pt",
+                voice_preset=voice_preset,
+            ).to("cuda")
+        else:
+            inputs = self.processor(
+                text=[text],
+                return_tensors="pt",
+            ).to("cuda")
 
         with torch.autocast("cuda"):
             outputs = self.model.generate(**inputs)

requirements.txt

@@ -1,2 +0,0 @@
-transformers==4.34.1
-accelerate>=0.23.0