Update app.py

app.py

@@ -1,38 +1,25 @@
-from transformers import pipeline
+import torch
+import torchaudio
 import gradio as gr
-import time
-import unicodedata
+from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 
-p = pipeline("automatic-speech-recognition", model="kingabzpro/wav2vec2-large-xls-r-300m-Urdu")
+# Load the custom model from Hugging Face Spaces
+model_name = "kingabzpro/wav2vec2-large-xls-r-300m-Urdu"
+model = Wav2Vec2ForCTC.from_pretrained(model_name)
+processor = Wav2Vec2Processor.from_pretrained(model_name)
 
+# Define the transcribe function
 def transcribe(audio):
-    text = p(audio)["text"]
-    transcription = unicodedata.normalize("NFC", text)
+    waveform, sample_rate = torchaudio.load(audio, normalize=True)
+    input_dict = processor(waveform, return_tensors="pt", padding=True)
+    logits = model(input_dict.input_values).logits
+    predicted_ids = torch.argmax(logits, dim=-1).squeeze()
+    transcription = processor.decode(predicted_ids)
     return transcription
 
-audio_input = gr.inputs.Audio(source="microphone", type="file", label="Upload audio file")
+# Define the interface
+audio_input = gr.inputs.Audio(source="microphone", type="numpy", label="Speak or Upload Audio")
 text_output = gr.outputs.Textbox(label="Transcription")
 
-interface = gr.Interface(
-    fn=transcribe,
-    inputs=audio_input,
-    outputs=text_output,
-    title="Real-Time Urdu ASR",
-    description="""
-    <p>
-    <center>
-    This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset.
-    </center>
-    </p>
-    <center>
-    <img src="https://huggingface.co/spaces/kingabzpro/real-time-Urdu-ASR/resolve/main/Images/cover.jpg" alt="logo" width="550"/>
-    </center>
-    """,
-    article="""
-    <p style='text-align: center'><a href='https://dagshub.com/kingabzpro/Urdu-ASR-SOTA' target='_blank'>Source Code on DagsHub</a></p><p style='text-align: center'><a href='https://huggingface.co/blog/fine-tune-xlsr-wav2vec2' target='_blank'>Fine-tuning XLS-R for Multi-Lingual ASR with 🤗 Transformers</a></p></center><center><img src='https://visitor-badge.glitch.me/badge?page_id=kingabzpro/real-time-Urdu-ASR' alt='visitor badge'></center></p>
-    """,
-    theme='EveryPizza/Cartoony-Gradio-Theme',
-    live=True
-)
-
+interface = gr.Interface(fn=transcribe, inputs=audio_input, outputs=text_output, title="Speech Recognition")
 interface.launch()