Update app.py

app.py

@@ -10,8 +10,23 @@ device = "cuda:0" if torch.cuda.is_available() else "cpu"
 model = CombinedModel("facebook/mms-1b-all", "Sunbird/sunbird-mul-en-mbart-merged", device=device)
 
 def transcribe(audio_file_mic=None, audio_file_upload=None):
-    # Your transcribe function code here
-    # ...
+    if audio_file_mic:
+        audio_file = audio_file_mic
+    elif audio_file_upload:
+        audio_file = audio_file_upload
+    else:
+        return "Please upload an audio file or record one"
+
+    # Make sure audio is 16kHz
+    speech, sample_rate = librosa.load(audio_file)
+    if sample_rate != 16000:
+        speech = librosa.resample(speech, orig_sr=sample_rate, target_sr=16000)
+    speech = torch.tensor([speech])
+
+    with torch.no_grad():
+        transcription, translation = model({"audio": speech})
+
+    return transcription, translation[0]
 
 description = '''Luganda to English Speech Translation'''