Update app.py

app.py

@@ -1,5 +1,5 @@
 import torch
-import torchaudio  # ✅ Added torchaudio to handle audio resampling
+import torchaudio  # ✅ Added torchaudio for resampling
 import gradio as gr
 import time  
 import numpy as np
@@ -11,9 +11,9 @@ device = "cpu"
 torch_dtype = torch.float32  # Use CPU-friendly float type
 MODEL_NAME = "openai/whisper-tiny"  # ✅ Switched to smallest model for fastest performance
 
-# ✅ 2️⃣ Load Whisper Tiny Model on CPU (Removed `low_cpu_mem_usage=True`)
+# ✅ 2️⃣ Load Whisper Tiny Model on CPU
 model = AutoModelForSpeechSeq2Seq.from_pretrained(
-    MODEL_NAME, torch_dtype=torch_dtype, use_safetensors=True  # ✅ Removed low_cpu_mem_usage
+    MODEL_NAME, torch_dtype=torch_dtype, use_safetensors=True
 )
 model.to(device)
 
@@ -28,7 +28,7 @@ pipe = pipeline(
     chunk_length_s=2,  # ✅ Process in 2-second chunks for ultra-low latency
     torch_dtype=torch_dtype,
     device=device,
-    sampling_rate=16000,  # ✅ Explicitly set sampling rate to avoid resampling issues
+    sampling_rate=16000,  # ✅ Explicitly set sampling rate
 )
 
 # ✅ 4️⃣ Real-Time Streaming Transcription (Microphone)
@@ -40,8 +40,99 @@ def stream_transcribe(stream, new_chunk):
         # ✅ Convert stereo to mono
         if y.ndim > 1:
             y = y.mean(axis=1)
-            
+
         y = y.astype(np.float32)
         y /= np.max(np.abs(y))
 
-        # ✅ Resample audio 
+        # ✅ Resample audio to 16kHz using torchaudio
+        y_tensor = torch.tensor(y)
+        resampler = torchaudio.transforms.Resample(orig_freq=sr, new_freq=16000)
+        y_resampled = resampler(y_tensor).numpy()
+        
+        # ✅ Append to Stream
+        if stream is not None:
+            stream = np.concatenate([stream, y_resampled])
+        else:
+            stream = y_resampled
+            
+        # ✅ Run Transcription
+        transcription = pipe({"sampling_rate": 16000, "raw": stream})["text"]
+        latency = time.time() - start_time
+
+        return stream, transcription, f"{latency:.2f} sec"
+
+    except Exception as e:
+        print(f"Error: {e}")
+        return stream, str(e), "Error"
+
+# ✅ 5️⃣ Transcription for File Upload
+def transcribe(inputs, previous_transcription):
+    start_time = time.time()
+    try:
+        # ✅ Convert file input to correct format
+        sample_rate, audio_data = inputs
+
+        # ✅ Resample to 16kHz using torchaudio
+        audio_tensor = torch.tensor(audio_data)
+        resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
+        resampled_audio = resampler(audio_tensor).numpy()
+
+        transcription = pipe({"sampling_rate": 16000, "raw": resampled_audio})["text"]
+
+        previous_transcription += transcription
+        latency = time.time() - start_time
+
+        return previous_transcription, f"{latency:.2f} sec"
+
+    except Exception as e:
+        print(f"Error: {e}")
+        return previous_transcription, "Error"
+
+# ✅ 6️⃣ Clear Function
+def clear():
+    return ""
+
+# ✅ 7️⃣ Gradio Interface (Microphone Streaming)
+with gr.Blocks() as microphone:
+    gr.Markdown(f"# Whisper Tiny - Real-Time Transcription (CPU) 🎙️")
+    gr.Markdown(f"Using [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) for ultra-fast speech-to-text.")
+
+    with gr.Row():
+        input_audio_microphone = gr.Audio(sources=["microphone"], type="numpy", streaming=True)
+        output = gr.Textbox(label="Live Transcription", value="")
+        latency_textbox = gr.Textbox(label="Latency (seconds)", value="0.0")
+
+    with gr.Row():
+        clear_button = gr.Button("Clear Output")
+
+    state = gr.State()
+    input_audio_microphone.stream(
+        stream_transcribe, [state, input_audio_microphone], 
+        [state, output, latency_textbox], time_limit=30, stream_every=1
+    )
+    clear_button.click(clear, outputs=[output])
+
+# ✅ 8️⃣ Gradio Interface (File Upload)
+with gr.Blocks() as file:
+    gr.Markdown(f"# Upload Audio File for Transcription 🎵")
+    gr.Markdown(f"Using [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) for speech-to-text.")
+
+    with gr.Row():
+        input_audio = gr.Audio(sources=["upload"], type="numpy")
+        output = gr.Textbox(label="Transcription", value="")
+        latency_textbox = gr.Textbox(label="Latency (seconds)", value="0.0")
+
+    with gr.Row():
+        submit_button = gr.Button("Submit")
+        clear_button = gr.Button("Clear Output")
+
+    submit_button.click(transcribe, [input_audio, output], [output, latency_textbox])
+    clear_button.click(clear, outputs=[output])
+
+# ✅ 9️⃣ Final Gradio App (Supports Microphone & File Upload)
+with gr.Blocks(theme=gr.themes.Ocean()) as demo:
+    gr.TabbedInterface([microphone, file], ["Microphone", "Upload Audio"])
+
+# ✅ 1️⃣0️⃣ Run Gradio Locally
+if __name__ == "__main__":
+    demo.launch()