Update app.py

app.py

@@ -1,32 +1,39 @@
 import gradio as gr
 import numpy as np
-from gradio_client import Client
+import torch
+from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
 
-client = Client("Pijush2023/voitex07122024")
+model_id = 'openai/whisper-large-v3'
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
+model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch_dtype).to(device)
+processor = AutoProcessor.from_pretrained(model_id)
+
+pipe_asr = pipeline("automatic-speech-recognition", model=model, tokenizer=processor.tokenizer, feature_extractor=processor.feature_extractor, max_new_tokens=128, chunk_length_s=15, batch_size=16, torch_dtype=torch_dtype, device=device, return_timestamps=True)
+
+def transcribe_function(new_chunk, state):
+    try:
+        sr, y = new_chunk[0], new_chunk[1]
+    except TypeError:
+        print(f"Error chunk structure: {type(new_chunk)}, content: {new_chunk}")
+        return state, "", None
+
+    y = y.astype(np.float32) / np.max(np.abs(y))
 
-def transcribe_audio_from_api(new_chunk, state):
-    sr, y = new_chunk
-    y_list = y.tolist()  # Convert NumPy array to list for JSON serialization
-    new_chunk_serialized = {"sampling_rate": sr, "array": y_list}
-    
-    # Update the state with the new chunk
     if state is not None:
-        state += new_chunk_serialized["array"]
+        state = np.concatenate([state, y])
     else:
-        state = new_chunk_serialized["array"]
-    
-    chunk_to_send = {"sampling_rate": sr, "array": state}
-    
-    result = client.predict(
-        new_chunk=chunk_to_send,
-        api_name="/SAMLOne_real_time"
-    )
-    
-    return state, result[1]  # Return the updated state and transcribed text
+        state = y
 
-with gr.Blocks() as frontend:
-    gr.Markdown("# Voice to Text Transcription (Frontend)")
+    result = pipe_asr({"array": state, "sampling_rate": sr}, return_timestamps=False)
 
+    full_text = result.get("text", "")
+    
+    return state, full_text
+
+with gr.Blocks() as demo:
+    gr.Markdown("# Voice to Text Transcription")
+    
     state = gr.State(None)
 
     with gr.Row():
@@ -35,6 +42,6 @@ with gr.Blocks() as frontend:
         with gr.Column():
             output_text = gr.Textbox(label="Transcription")
 
-    audio_input.stream(transcribe_audio_from_api, inputs=[audio_input, state], outputs=[state, output_text])
+    audio_input.stream(transcribe_function, inputs=[audio_input, state], outputs=[state, output_text], api_name="SAMLOne_real_time")
 
-frontend.launch()
+demo.launch(show_error=True)