Sync with https://github.com/mozilla-ai/speech-to-text-finetune

app.py

@@ -53,32 +53,26 @@ def _load_hf_model(model_repo_id: str) -> Tuple[Pipeline | None, str]:
     return pipeline(
         "automatic-speech-recognition",
         model=model_repo_id,
-        device=0
     ), f"✅ HF Model {model_repo_id} has been loaded."
 
 
-def load_model(
-    dropdown_model_id: str, hf_model_id: str, local_model_id: str
-) -> Tuple[Pipeline, str]:
+@spaces.GPU
+def transcribe(
+    dropdown_model_id: str,
+    hf_model_id: str,
+    local_model_id: str,
+    pipe: Pipeline,
+    audio: gr.Audio,
+) -> str:
     if dropdown_model_id and not hf_model_id and not local_model_id:
         dropdown_model_id = dropdown_model_id.split(" (")[0]
-        yield None, f"Loading {dropdown_model_id}..."
-        yield _load_hf_model(dropdown_model_id)
+        pipe = _load_hf_model(dropdown_model_id)
     elif hf_model_id and not local_model_id and not dropdown_model_id:
-        yield None, f"Loading {hf_model_id}..."
-        yield _load_hf_model(hf_model_id)
+        pipe = _load_hf_model(hf_model_id)
     elif local_model_id and not hf_model_id and not dropdown_model_id:
-        yield None, f"Loading {local_model_id}..."
-        yield _load_local_model(local_model_id)
+        pipe = _load_local_model(local_model_id)
     else:
-        yield (
-            None,
-            "️️⚠️ Please select or fill at least and only one of the options above",
-        )
-
-
-@spaces.GPU
-def transcribe(pipe: Pipeline, audio: gr.Audio) -> str:
+        return ("️️⚠️ Please select or fill at least and only one of the options above",)
     text = pipe(audio)["text"]
     return text
 
@@ -111,7 +105,7 @@ def setup_gradio_demo():
                     placeholder="artifacts/my-whisper-tiny",
                 )
 
-        load_model_button = gr.Button("Load model")
+        # load_model_button = gr.Button("Load model")
         model_loaded = gr.Markdown()
 
         ### Transcription ###
@@ -126,15 +120,19 @@ def setup_gradio_demo():
         transcribe_output = gr.Text(label="Output")
 
         ### Event listeners ###
+        """
         model = gr.State()
         load_model_button.click(
             fn=load_model,
             inputs=[dropdown_model, user_model, local_model],
             outputs=[model, model_loaded],
         )
+        """
 
         transcribe_button.click(
-            fn=transcribe, inputs=[model, audio_input], outputs=transcribe_output
+            fn=transcribe,
+            inputs=[dropdown_model, user_model, local_model, audio_input],
+            outputs=transcribe_output,
         )
 
     demo.launch()