accepting input from upload or mic

app.py

@@ -3,13 +3,21 @@ os.system("pip install git+https://github.com/openai/whisper.git")
 import gradio as gr
 import whisper
 
-from transformers import WhisperProcessor, WhisperForConditionalGeneration
-
 model = whisper.load_model("small")
 
+model.config.forced_decoder_ids = None
+
 
-def inference(audio):
-    audio = whisper.load_audio(audio)
+def predict(audio, mic_audio=None):
+    # audio = tuple (sample_rate, frames) or (sample_rate, (frames, channels))
+    if mic_audio is not None:
+        sampling_rate, waveform = mic_audio
+    elif audio is not None:
+        sampling_rate, waveform = audio
+    else:
+        return "(please provide audio)"
+
+    audio = whisper.load_audio(waveform)
     audio = whisper.pad_or_trim(audio)
     
     mel = whisper.log_mel_spectrogram(audio).to(model.device)
@@ -23,6 +31,7 @@ def inference(audio):
     return result.text, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
 
 
+
 title = "Demo for Whisper -> Something -> XLS-R"
 
 description = """
@@ -31,7 +40,7 @@ being passed into the model. The output is the text transcription of the audio.
 """
 
 gr.Interface(
-    fn=inference,
+    fn=predict,
     inputs=[
         gr.Audio(label="Upload Speech", source="upload", type="numpy"),
         gr.Audio(label="Record Speech", source="microphone", type="numpy"),