Update app.py

app.py

@@ -1,52 +1,52 @@
-#Importing all the necessary packages
+# Importing all the necessary packages
 import nltk
-import librosa
 import torch
 import gradio as gr
-from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
+from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
+import numpy as np
+
+# Downloading the necessary NLTK data
 nltk.download("punkt")
 
-#Loading the pre-trained model and the tokenizer
+# Loading the pre-trained model and the processor
 model_name = "facebook/wav2vec2-base-960h"
-tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
+processor = Wav2Vec2Processor.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.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker"),
-             outputs = gr.outputs.Textbox(label="Output Text"),
-             title="ASR using Wav2Vec 2.0",
-             description = "This application displays transcribed text for given audio input",
-             examples = [["Test_File1.wav"], ["Test_File2.wav"], ["Test_File3.wav"]], theme="grass").launch()
-
+    sentences = nltk.sent_tokenize(input_sentence)
+    return ' '.join([s.replace(s[0], s[0].capitalize(), 1) for s in sentences])
+
+def asr_transcript(audio):
+    # Process the audio
+    input_values = processor(audio, sampling_rate=16000, return_tensors="pt").input_values
+    # Get logits
+    logits = model(input_values).logits
+    # Get predicted IDs
+    predicted_ids = torch.argmax(logits, dim=-1)
+    # Decode the IDs to text
+    transcription = processor.decode(predicted_ids[0])
+    # Correct the casing
+    transcription = correct_casing(transcription.lower())
+    return transcription
+
+def real_time_asr(audio, state=None):
+    if state is None:
+        state = ""
+    audio = np.array(audio)
+    transcription = asr_transcript(audio)
+    state += " " + transcription
+    return state, state
+
+# Create the Gradio interface
+iface = gr.Interface(
+    fn=real_time_asr,
+    inputs=[gr.Audio(source="microphone", streaming=True), "state"],
+    outputs="text",
+    live=True,
+    title="Real-Time ASR using Wav2Vec 2.0",
+    description="This application displays transcribed text in real-time for given audio input"
+)
+
+# Launch the interface
+iface.launch()