Create app.py

app.py

@@ -0,0 +1,31 @@
+import gradio as gr
+import torch
+from espnet2.bin.asr_inference import Speech2Text
+from espnet_model_zoo.downloader import ModelDownloader
+
+# Download a pretrained model
+d = ModelDownloader()
+asr_model = Speech2Text(
+    **d.download_and_unpack("espnet/simple_asr_train_asr_transformer_e18_raw_bpe_sp_valid.acc.best"),
+    device="cpu",  # Change to "cuda" if using a GPU
+)
+
+def transcribe(audio):
+    """Transcribe speech to text using ESPnet."""
+    # Convert audio input (from Gradio) to text
+    speech = torch.tensor(audio[1])  # Extract the audio waveform
+    result = asr_model(speech)
+    text, *_ = result[0]  # Get the transcription from the result
+    return text
+
+# Create a simple Gradio interface
+interface = gr.Interface(
+    fn=transcribe,                   # Function to call
+    inputs=gr.Audio(source="microphone", type="numpy"),  # Audio input from microphone
+    outputs="text",                  # Output type (text transcription)
+    title="ESPnet ASR Demo",         # Title of the UI
+    description="Simple ESPnet-based speech recognition",  # Description of the app
+)
+
+# Launch the app
+interface.launch()