Update app.py

app.py

@@ -1,41 +1,47 @@
 import gradio as gr
 import torch
 import torchaudio
+import numpy as np
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 
-# Load the Swahili ASR model
+# Load your trained Swahili ASR model
 model_name = "Futuresony/Future-sw_ASR-24-02-2025"
 processor = Wav2Vec2Processor.from_pretrained(model_name)
 model = Wav2Vec2ForCTC.from_pretrained(model_name)
 
-# Function to process live audio stream
-def transcribe_live(microphone_audio):
-    speech_array, sample_rate = torchaudio.load(microphone_audio)
+# Process microphone input in real-time
+def transcribe_live(audio):
+    if audio is None:
+        return ""
     
-    # Resample to 16kHz
-    resampler = torchaudio.transforms.Resample(orig_freq=sample_rate, new_freq=16000)
-    speech_array = resampler(speech_array).squeeze().numpy()
+    # Convert NumPy array to PyTorch tensor
+    speech_array = torch.from_numpy(audio).float()
     
-    # Process and transcribe
-    input_values = processor(speech_array, sampling_rate=16000, return_tensors="pt").input_values
+    # Resample audio to 16kHz (if needed)
+    sample_rate = 16000  # Since streaming provides 16kHz by default
+    
+    # Process input
+    input_values = processor(speech_array, sampling_rate=sample_rate, return_tensors="pt").input_values
+
+    # Predict transcription
     with torch.no_grad():
         logits = model(input_values).logits
     predicted_ids = torch.argmax(logits, dim=-1)
-    
-    # Decode the text
+
+    # Decode text
     transcription = processor.batch_decode(predicted_ids)[0]
     return transcription
 
-# Create Gradio interface with live microphone input
+# Create Gradio interface with real-time streaming
 interface = gr.Interface(
     fn=transcribe_live,
-    inputs=gr.Audio(sources=["microphone"], type="filepath"),
-    outputs="text",
+    inputs=gr.Audio(streaming=True, type="numpy"),  # Live streaming input
+    outputs=gr.Textbox(label="Live Transcription"),
     live=True,
-    title="Live Swahili ASR Transcription",
-    description="Speak into your microphone, and the model will transcribe in real-time.",
+    title="Live Swahili ASR Streaming",
+    description="Talk and see real-time Swahili subtitles appear below!",
 )
 
-# Launch the app
+# Launch the live streaming ASR app
 if __name__ == "__main__":
     interface.launch()