testing multiple models

app.py

@@ -3,14 +3,13 @@ import librosa
 import torch
 import gradio as gr
 from pyctcdecode import build_ctcdecoder
-from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
+from transformers import AutoModelProcessor, AutoModelForCTC
 
 nltk.download("punkt")
 
-#Loading the model and the tokenizer
 model_name = "facebook/wav2vec2-base-960h"
-processor = Wav2Vec2Processor.from_pretrained(model_name)
-model = Wav2Vec2ForCTC.from_pretrained(model_name)
+processor = AutoModelProcessor.from_pretrained(model_name)
+model = AutoModelForCTC.from_pretrained(model_name)
 
 def load_and_fix_data(input_file):  
   #read the file
@@ -59,7 +58,7 @@ def return_all_predictions(input_file):
 
 
 gr.Interface(return_all_predictions,
-             inputs = gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Record/ Drop audio"),
+             inputs = [gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Record/ Drop audio"), gr.inputs.Dropdown(["facebook/wav2vec2-base-960h", "facebook/hubert-large-ls960-ft"])],
              outputs = [gr.outputs.Textbox(label="Beam CTC Decoding"), gr.outputs.Textbox(label="Greedy Decoding")],
              title="ASR using Wav2Vec 2.0 & pyctcdecode",
              description = "Extending HF ASR models with pyctcdecode decoder",