app.py

app.py

@@ -0,0 +1,55 @@
+#Importing all the necessary packages
+import nltk
+import librosa
+import torch
+import gradio as gr
+from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
+nltk.download("punkt")
+
+#Loading the pre-trained model and the tokenizer
+model_name = "shizukanabasho/north2"
+tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
+model = Wav2Vec2ForCTC.from_pretrained(model_name)
+
+def load_data(input_file):
+
+  #reading the file
+  speech, sample_rate = librosa.load(input_file)
+  #make it 1-D
+  if len(speech.shape) > 1: 
+      speech = speech[:,0] + speech[:,1]
+  #Resampling the audio at 16KHz
+  if sample_rate !=16000:
+    speech = librosa.resample(speech, sample_rate,16000)
+  return speech
+  
+def correct_casing(input_sentence):
+
+  sentences = nltk.sent_tokenize(input_sentence)
+  return (''.join([s.replace(s[0],s[0].capitalize(),1) for s in sentences]))
+  
+def asr_transcript(input_file):
+
+  speech = load_data(input_file)
+  #Tokenize
+  input_values = tokenizer(speech, return_tensors="pt").input_values
+  #Take logits
+  logits = model(input_values).logits
+  #Take argmax
+  predicted_ids = torch.argmax(logits, dim=-1)
+  #Get the words from predicted word ids
+  transcription = tokenizer.decode(predicted_ids[0])
+  #Correcting the letter casing
+  # transcription = correct_casing(transcription.lower())
+  return transcription
+
+gr.Interface(
+    asr_transcript,
+    inputs=[
+        gr.Audio(source="microphone", type="filepath", optional=True),
+        gr.Audio(source="upload", type="filepath", optional=True),
+        title="ASR using Wav2Vec2.0",
+        description = "This application displays transcribed text for given audio input",
+    ],
+    outputs="text",
+).launch()