Update app.py

app.py

@@ -1,13 +1,20 @@
 import os
 import gradio as gr
 import numpy as np
-
+from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
+from fairseq.models.speech_to_speech.hub_interface import S2SHubInterface
+from fairseq.models.speech_to_text.hub_interface import S2THubInterface
+form api-inference-community.docker_images.fairseq.app.pipelines.audio_to_audio import SpeechToSpeechPipeline
 
 io1 = gr.Interface.load("huggingface/facebook/xm_transformer_s2ut_en-hk", api_key=os.environ['api_key'])
 io2 = gr.Interface.load("huggingface/facebook/xm_transformer_s2ut_hk-en", api_key=os.environ['api_key'])
 io3 = gr.Interface.load("huggingface/facebook/xm_transformer_unity_en-hk", api_key=os.environ['api_key'])
 io4 = gr.Interface.load("huggingface/facebook/xm_transformer_unity_hk-en", api_key=os.environ['api_key'])
-   
+
+def call_model(audio, model):
+    pipe = SpeechToSpeechPipeline("facebook/xm_transformer_unity_hk-en")
+    return pipe(audio)
+
 def inference(audio, model):
     if model == "xm_transformer_s2ut_en-hk":
         out_audio = io1(audio)
@@ -15,6 +22,8 @@ def inference(audio, model):
         out_audio = io2(audio)
     elif model == "xm_transformer_unity_en-hk":
         out_audio = io3(audio)
+    elif model == "xm_transformer_unity_hk-en_gpu":
+        out_audio = call_model(audio, model)
     else:
         out_audio = io4(audio)
     return out_audio 
@@ -128,7 +137,7 @@ with block:
                 )
                 
                 btn = gr.Button("Submit")
-        model = gr.Dropdown(choices=["xm_transformer_unity_en-hk", "xm_transformer_unity_hk-en", "xm_transformer_s2ut_en-hk", "xm_transformer_s2ut_hk-en"], value="xm_transformer_unity_en-hk",type="value", label="Model")
+        model = gr.Dropdown(choices=["xm_transformer_unity_en-hk", "xm_transformer_unity_hk-en", "xm_transformer_unity_hk-en_gpu", "xm_transformer_s2ut_en-hk", "xm_transformer_s2ut_hk-en"], value="xm_transformer_unity_en-hk",type="value", label="Model")
         out = gr.Audio(label="Output")
         
         btn.click(inference, inputs=[audio, model], outputs=out)