Update app.py

app.py

@@ -85,7 +85,11 @@ from transformers import pipeline
 import librosa
 import numpy as np
 
-transcriber = pipeline("automatic-speech-recognition", model="Rezuwan/regional_asr_weights")
+
+MODEL_NAME = "Rezuwan/regional_asr_weights"
+
+
+transcriber = pipeline("automatic-speech-recognition", model=MODEL_NAME)
 
 def transcribe_audio(audio_path):
     try:
@@ -102,8 +106,30 @@ iface = gr.Interface(
     fn=transcribe_audio,
     inputs=gr.Audio(type="filepath", label="Upload or Record Audio"),
     outputs=gr.Textbox(label="Transcription"),
-    title="🗣️ Barishal ASR Speech-to-Text",
-    description="Upload or record an audio file to transcribe Bengali speech using the Barishal ASR model."
+    title="Bengali Speech-to-Text with Regional Dialects",
+    description=(
+        f"""
+        Model Card: [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files of arbitrary length. [Do leave a like (❤️) on the model card and this space]
+        
+        Instructions:
+        
+        1. Click on 'Record' option in the left 'Upload or Record Audio' section and record the audio. 
+        2. When done recording, click on 'Stop' button and give it some time until some waveform shows up in the 'Upload or Record Audio' section (Same goes when uploading pre-recorded audio files) and then click the 'Submit' button. 
+        3. Wait for the audio clip to be processed (This could take a while 😅. Still needs work on the inference time) and then transcription of the audio will appear on the right 'output' section.
+        4. If want to submit a trimmed version of the input, select the trimmed audio snippet and then click 'Trim' and then wait a bit until wavform 
+        shows up in the input section of the interface and then click 'Submit'.
+
+
+
+        Note: 
+        
+        1. Since the corpus used to fine-tune this model was really small, The orthography might still not be upto the mark but it gets the work done but still needs work and manual validation.
+        
+        2.With proper data and a larger version of the corpus, I guess I'll be able to increase it's transcription performance of the Bengali speech with regional dialects.
+
+        ![](screenshot.jpg)
+        """
+    )
 )
 
 iface.launch()