Fixed audio to audio & better error messaging if Space is not loading

gradio/external.py

@@ -2,6 +2,7 @@ import base64
 import json
 import re
 import tempfile
+from pydantic import MissingError
 
 import requests
 
@@ -24,13 +25,27 @@ def get_huggingface_interface(model_name, api_key, alias):
     p = response.json().get("pipeline_tag")
 
     def encode_to_base64(r: requests.Response) -> str:
+        # Handles the different ways HF API returns the prediction
         base64_repr = base64.b64encode(r.content).decode("utf-8")
         data_prefix = ";base64,"
+        # Case 1: base64 representation already includes data prefix
         if data_prefix in base64_repr:
             return base64_repr
         else:
             content_type = r.headers.get("content-type")
-            return "data:{};base64,".format(content_type) + base64_repr
+            # Case 2: the data prefix is a key in the response
+            if content_type == "application/json":
+                try: 
+                    content_type = r.json()[0]["content-type"]
+                    base64_repr = r.json()[0]["blob"]
+                except KeyError:
+                    raise ValueError("Cannot determine content type returned"
+                                     "by external API.")
+            # Case 3: the data prefix is included in the response headers
+            else:
+                pass
+            new_base64 = "data:{};base64,".format(content_type) + base64_repr
+            return new_base64
 
     pipelines = {
         "audio-classification": {
@@ -45,7 +60,7 @@ def get_huggingface_interface(model_name, api_key, alias):
             },
         },
         "audio-to-audio": {
-            # example model: https://hf.co/ehcalabres/wav2vec2-lg-xlsr-en-speech-emotion-recognition
+            # example model: https://hf.co/speechbrain/mtl-mimic-voicebank
             "inputs": inputs.Audio(label="Input", source="upload", type="filepath"),
             "outputs": outputs.Audio(label="Output"),
             "preprocess": lambda i: base64.b64decode(
@@ -192,7 +207,7 @@ def get_huggingface_interface(model_name, api_key, alias):
     }
 
     if p is None or not (p in pipelines):
-        raise ValueError("Unsupported pipeline type: {}".format(type(p)))
+        raise ValueError("Unsupported pipeline type: {}".format(p))
 
     pipeline = pipelines[p]
 
@@ -282,7 +297,10 @@ def get_spaces_interface(model_name, api_key, alias):
     result = re.search(
         "window.gradio_config = (.*?);</script>", r.text
     )  # some basic regex to extract the config
-    config = json.loads(result.group(1))
+    try:
+        config = json.loads(result.group(1))
+    except AttributeError:
+        raise ValueError("Could not load the Space: {}".format(model_name))
     interface_info = interface_params_from_config(config)
 
     # The function should call the API with preprocessed data

test/test_external.py

@@ -19,7 +19,7 @@ class TestHuggingFaceModelAPI(unittest.TestCase):
     def test_audio_to_audio(self):
         model_type = "audio-to-audio"
         interface_info = gr.external.get_huggingface_interface(
-            "facebook/xm_transformer_600m-es_en-multi_domain",
+            "speechbrain/mtl-mimic-voicebank",
             api_key=None,
             alias=model_type,
         )