Update app.py

app.py

@@ -1,25 +1,77 @@
-import torch
-import torchaudio
+import os
+import unicodedata
+from datasets import load_dataset, Audio
+from transformers import pipeline
 import gradio as gr
-from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-
-# 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):
-    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
-
-# 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="Speech Recognition")
-interface.launch()
+import torch
+
+############### HF ###########################
+
+HF_TOKEN = os.getenv("HF_TOKEN")
+
+hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "Urdu-ASR-flags")
+
+############## DagsHub ################################
+
+Model = "kingabzpro/wav2vec2-large-xls-r-300m-Urdu"
+# This is not working because Huggingface has completely changed the git server. 
+# from dagshub.streaming import install_hooks
+# install_hooks()
+
+############## Inference ##############################
+
+
+def asr(audio):
+
+    asr = pipeline("automatic-speech-recognition", model=Model)
+    prediction = asr(audio, chunk_length_s=30)
+    return unicodedata.normalize("NFC",prediction["text"])
+
+
+################### Gradio Web APP ################################
+
+title = "Urdu Automatic Speech Recognition"
+
+description = """
+<p>
+<center>
+This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset.
+<img src="https://huggingface.co/spaces/kingabzpro/Urdu-ASR-SOTA/resolve/main/Images/cover.jpg" alt="logo" width="550"/>
+</center>
+</p>
+"""
+
+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/Urdu-ASR-SOTA' alt='visitor badge'></center></p>"
+
+examples = [["Sample/sample1.mp3"], ["Sample/sample2.mp3"], ["Sample/sample3.mp3"]]
+
+
+Input = gr.Audio(
+    source="microphone",
+    type="filepath",
+    label="Please Record Your Voice",
+)
+Output = gr.Textbox(label="Urdu Script")
+
+
+def main():
+    iface = gr.Interface(
+        asr,
+        Input,
+        Output,
+        title=title,
+        allow_flagging="manual",
+        flagging_callback=hf_writer,
+        description=description,
+        article=article,
+        examples=examples,
+        theme='JohnSmith9982/small_and_pretty'
+    )
+
+    iface.launch(enable_queue=True)
+
+
+# enable_queue=True,auth=("admin", "pass1234")
+
+if __name__ == "__main__":
+    main()