Update app.py

app.py

@@ -3,7 +3,6 @@ import numpy as np
 import torch
 from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
 
-# Initialize the model and processor
 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
@@ -12,24 +11,37 @@ 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(audio):
-    sr, y = audio
+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))
-    result = pipe_asr({"array": y, "sampling_rate": sr}, return_timestamps=False)
+
+    if state is not None:
+        state = np.concatenate([state, y])
+    else:
+        state = y
+
+    result = pipe_asr({"array": state, "sampling_rate": sr}, return_timestamps=False)
+
     full_text = result.get("text", "")
-    return full_text
+    
+    return state, full_text
 
 with gr.Blocks() as demo:
     gr.Markdown("# Voice to Text Transcription\nTranscribe your voice input to text using a pre-trained Whisper model.")
     
+    state = gr.State(None)
+
     with gr.Row():
         with gr.Column():
-            audio_input = gr.Audio(sources="microphone", type="numpy", label="Speak Here")
-            transcribe_button = gr.Button("Transcribe")
-        
+            audio_input = gr.Audio(sources=["microphone"], streaming=True, type='numpy', label="Microphone Input")
         with gr.Column():
             output_text = gr.Textbox(label="Transcription")
 
-    transcribe_button.click(transcribe_function, inputs=audio_input, outputs=output_text)
+    audio_input.stream(transcribe_function, inputs=[audio_input, state], outputs=[state, output_text], api_name="SAMLOne_real_time")
 
 demo.launch()