working

handler.py

@@ -11,7 +11,7 @@ class EndpointHandler():
     def __init__(self, path=""):
         # load the optimized model
         # create inference pipeline
-        self.pipeline = pipeline("text-to-audio", "facebook/musicgen-stereo-large", device="cuda", torch_dtype=torch.float16)
+        self.pipeline = pipeline("text-to-audio", "facebook/musicgen-stereo-large", device="mps", torch_dtype=torch.float16)
 
     def generate_audio(self, text: str):
         # Here you can implement your audio generation logic
@@ -30,13 +30,9 @@ class EndpointHandler():
 
         audio_data, sampling_rate = self.generate_audio(input)
 
-        
-        # Convert audio data to base64 string
-        audio_base64 = base64.b64encode(audio_data.tobytes())
-
         # Create JSON response
         response = {
-            "audio_base64": audio_base64,
+            "audio_data": audio_data,
             "sampling_rate": sampling_rate
         }
 

test.py

@@ -1,4 +1,7 @@
 from handler import EndpointHandler
+import soundfile as sf
+import numpy as np
+
 
 # init handler
 my_handler = EndpointHandler(path=".")
@@ -10,5 +13,11 @@ payload = {"inputs": "Lowfi hiphop with deep bass"}
 pred=my_handler(payload)
 
 
-with open("generated_audio.wav", "wb") as f:
-    f.write(pred)
+audio_data = pred["audio_data"]
+sampling_rate = pred["sampling_rate"]
+
+# Write the audio data to a WAV file
+output_file_path = "output.wav"  # Specify the file path
+sf.write(output_file_path, audio_data, sampling_rate)
+
+print("Audio file saved successfully.")

test_api.py

@@ -1,8 +1,9 @@
 import requests
 import base64
 import soundfile as sf
+import numpy as np
 
-API_URL = "https://ooj7l6a226z6pn8j.us-east-1.aws.endpoints.huggingface.cloud"
+API_URL = "https://gij43le0roc2pmst.us-east-1.aws.endpoints.huggingface.cloud"
 headers = {
 	"Accept" : "application/json",
 	"Authorization": "Bearer token",
@@ -18,18 +19,22 @@ response = query({
 	"parameters": {}
 })
 
+print(response)
+
 # Extract audio data and sampling rate from the JSON response
 audio_base64 = response["audio_base64"]
 sampling_rate = response["sampling_rate"]
 
 # Decode the base64-encoded audio data
 audio_binary = base64.b64decode(audio_base64)
+# Convert binary audio data to a NumPy array
+audio_np = np.frombuffer(audio_binary, dtype=np.int16)
+
+# Print the shape of the audio data
+print("Shape of audio data:", audio_np.shape)
 
 # Write the audio data to a WAV file
 output_file_path = "output.wav"  # Specify the file path
-sf.write(output_file_path, audio_binary, sampling_rate)
+sf.write(output_file_path, audio_np, sampling_rate)
 
-# Write the audio data to a file
-output_file_path = "output.wav"  # Specify the file path
-with open(output_file_path, "wb") as f:
-    f.write(audio_binary)
+print("Audio file saved successfully.")