Update app.py

app.py

@@ -1,64 +1,67 @@
+import torch
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import WhisperProcessor, WhisperForConditionalGeneration
+import librosa
+import numpy as np
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+# Check if GPU is available
+device = "cuda" if torch.cuda.is_available() else "cpu"
+print(f"Using device: {device}")
 
+# Load Whisper Model
+model_name = "openai/whisper-small"
+processor = WhisperProcessor.from_pretrained(model_name)
+model = WhisperForConditionalGeneration.from_pretrained(model_name, torch_dtype=torch.float16).to(device)
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+# Set the language token for Urdu transcription
+forced_decoder_ids = processor.get_decoder_prompt_ids(language="urdu", task="transcribe")
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+# Function to transcribe audio
+def transcribe_audio(audio_input):
+    if isinstance(audio_input, tuple):
+        audio, sr = audio_input
+    else:
+        audio, sr = librosa.load(audio_input, sr=16000)
 
-    messages.append({"role": "user", "content": message})
+    # Convert stereo to mono if necessary
+    if len(audio.shape) > 1:
+        audio = np.mean(audio, axis=0)
 
-    response = ""
+    # Resample if needed
+    if sr != 16000:
+        audio = librosa.resample(y=audio, orig_sr=sr, target_sr=16000)
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+    # Process audio
+    input_features = processor(
+        audio,
+        sampling_rate=16000,
+        return_tensors="pt",
+        padding=True,
+        return_attention_mask=True
+    ).to(device)
 
-        response += token
-        yield response
+    # Generate transcription
+    with torch.no_grad():
+        predicted_ids = model.generate(
+            input_features.input_features,
+            attention_mask=input_features.attention_mask,
+            forced_decoder_ids=forced_decoder_ids
+        )
 
+    # Decode transcription
+    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
+    return transcription
 
+# Create Gradio interface
+interface = gr.Interface(
+    fn=transcribe_audio,
+    inputs=gr.Audio(label="Record or Upload Audio", type="numpy"),
+    outputs=gr.Textbox(label="Transcription"),
+    title="Urdu Speech-to-Text Converter",
+    description="Record or upload Urdu speech to convert it to text using the open-source Whisper model."
+)
 
+# Launch Gradio app
 if __name__ == "__main__":
-    demo.launch()
+    interface.launch()