Update app.py

app.py

@@ -1,6 +1,9 @@
 import warnings
 import gradio as gr
 from transformers import pipeline
+from transformers import AutoProcessor
+from pyctcdecode import build_ctcdecoder
+from transformers import Wav2Vec2ProcessorWithLM
 
 import os
 import re
@@ -12,8 +15,24 @@ p2 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-bert-2.0-h
 #p3 = pipeline(task="automatic-speech-recognition", model="cdactvm/kannada_w2v-bert_model")
 #p4 = pipeline(task="automatic-speech-recognition", model="cdactvm/telugu_w2v-bert_model")
 #p5 = pipeline(task="automatic-speech-recognition", model="Sajjo/w2v-bert-2.0-bangala-gpu-CV16.0_v2")
-p6 = pipeline(task="automatic-speech-recognition", model="cdactvm/hf-open-assames")
+#p6 = pipeline(task="automatic-speech-recognition", model="cdactvm/hf-open-assames")
 p7 = pipeline(task="automatic-speech-recognition", model="cdactvm/w2v-assames")
+processor = AutoProcessor.from_pretrained("cdactvm/w2v-assames")
+vocab_dict = processor.tokenizer.get_vocab()
+sorted_vocab_dict = {k.lower(): v for k, v in sorted(vocab_dict.items(), key=lambda item: item[1])}
+decoder = build_ctcdecoder(
+    labels=list(sorted_vocab_dict.keys()),
+    kenlm_model_path="lm.binary",
+    )
+processor_with_lm = Wav2Vec2ProcessorWithLM(
+    feature_extractor=processor.feature_extractor,
+    tokenizer=processor.tokenizer,
+    decoder=decoder
+    )
+processor.feature_extractor._processor_class = "Wav2Vec2ProcessorWithLM"
+p8 = pipeline("automatic-speech-recognition", model="cdactvm/w2v-assames", tokenizer=processor_with_lm, feature_extractor=processor_with_lm.feature_extractor, decoder=processor_with_lm.decoder)
+    
+
 os.system('git clone https://github.com/irshadbhat/indic-trans.git')
 os.system('pip install ./indic-trans/.')
 
@@ -55,8 +74,8 @@ def transcribe_bangala(speech):
         return "Error: ASR returned None"
     return text
 
-def transcribe_assamese_model1(speech):
-    text = p6(speech)["text"]
+def transcribe_assamese_LM(speech):
+    text = p8(speech)["text"]
     text = cleanhtml(text)
     if text is None:
         return "Error: ASR returned None"
@@ -174,8 +193,8 @@ def sel_lng(lng, mic=None, file=None):
        return transcribe_ban_eng(audio)
     elif lng == "Bangala":
        return transcribe_bangala(audio)
-    elif lng == "Assamese-Model1":
-       return transcribe_assamese_model1(audio)
+    elif lng == "Assamese-LM":
+       return transcribe_assamese_LM(audio)
     elif lng == "Assamese-Model2":
        return transcribe_assamese_model2(audio)
         
@@ -404,7 +423,7 @@ demo=gr.Interface(
     inputs=[
         
         #gr.Dropdown(["Hindi","Hindi-trans","Odiya","Odiya-trans","Kannada","Kannada-trans","Telugu","Telugu-trans","Bangala","Bangala-trans"],value="Hindi",label="Select Language"),
-        gr.Dropdown(["Hindi","Hindi-trans","Assamese-Model1","Assamese-Model2"],value="Hindi",label="Select Language"),
+        gr.Dropdown(["Hindi","Hindi-trans","Assamese-LM","Assamese-Model2"],value="Hindi",label="Select Language"),
         gr.Audio(sources=["microphone","upload"], type="filepath"),
         #gr.Audio(sources="upload", type="filepath"),
         #"state"