Update app.py

app.py

@@ -15,22 +15,24 @@ auth_token = os.environ.get("HF_TOKEN")
 
 
 target_lang_options = {"English": "eng", "Luganda": "lug", "Acholi": "ach", "Runyankole": "nyn", "Lugbara": "lgg"}
-target_lang_code = target_lang_options[target_lang]
 
 languages = list(target_lang_options.keys())
 
 
-if target_lang_code=="eng":
-    model_id = "facebook/mms-1b-all"
-else:
-    model_id = "Sunbird/sunbird-mms"
-
-
 # Transcribe audio using custom model
-def transcribe_audio(input_file, target_lang_code,
+def transcribe_audio(input_file, language,
                       device, model_id=model_id, 
                       chunk_length_s=10, stride_length_s=(4, 2), return_timestamps="word"):
     
+
+    target_lang_code = target_lang_options[target_lang_code]
+    
+    # Determine the model_id based on the language
+    if target_lang_code == "eng":
+        model_id = "facebook/mms-1b-all"
+    else:
+        model_id = "Sunbird/sunbird-mms"
+                
     pipe = pipeline(model=model_id, device=device, token=hf_auth_token)  
     pipe.tokenizer.set_target_lang(target_lang_code)
     pipe.model.load_adapter(target_lang_code)
@@ -41,41 +43,13 @@ def transcribe_audio(input_file, target_lang_code,
     return output
 
 
-# def transcribe(audio_file_mic=None, audio_file_upload=None, language="Luganda (lug)"):
-#     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)
-
-#     # Keep the same model in memory and simply switch out the language adapters by calling load_adapter() for the model and set_target_lang() for the tokenizer
-#     language_code = language
-#     processor.tokenizer.set_target_lang(language_code)
-#     model.load_adapter(language_code)
-
-#     inputs = processor(speech, sampling_rate=16_000, return_tensors="pt")
-
-#     with torch.no_grad():
-#         outputs = model(**inputs).logits
-
-#     ids = torch.argmax(outputs, dim=-1)[0]
-#     transcription = processor.decode(ids)
-#     return transcription
-
-
 description = '''ASR with salt-mms'''
 
 iface = gr.Interface(fn=transcribe_audio,
                      inputs=[
                          gr.Audio(source="microphone", type="filepath", label="Record Audio"),
                          gr.Audio(source="upload", type="filepath", label="Upload Audio"),
-                         gr.Dropdown(choices=languages, label="Language", value="lug")
+                         gr.Dropdown(choices=languages, label="Language", value="English"
                          ],
                      outputs=gr.Textbox(label="Transcription"),
                      description=description