Upload app.py with huggingface_hub

app.py

@@ -9,8 +9,6 @@ import os
 import sys
 import pickle
 import tempfile
-import numpy as np
-import scipy.io.wavfile as wavfile
 from huggingface_hub import hf_hub_download
 
 # Add text2midi model to path
@@ -126,24 +124,25 @@ def midi_to_wav(midi_path: str, wav_path: str, sample_rate: int = 44100) -> bool
     return os.path.exists(wav_path)
 
 def generate_music(prompt: str):
-    """Generate music from text prompt. Returns audio as base64-encoded dict for API compatibility."""
+    """Generate music from text prompt. Returns audio file path."""
     if not prompt or not prompt.strip():
         return None
 
     midi_path = None
-    wav_path = None
 
     try:
-        # Create temporary files
+        # Create temporary MIDI file
         midi_fd, midi_path = tempfile.mkstemp(suffix='.mid')
         os.close(midi_fd)
 
-        wav_fd, wav_path = tempfile.mkstemp(suffix='.wav')
-        os.close(wav_fd)
+        # Create WAV in Gradio's temp directory for proper file serving
+        gradio_temp = os.path.join(tempfile.gettempdir(), "gradio")
+        os.makedirs(gradio_temp, exist_ok=True)
+        wav_path = os.path.join(gradio_temp, f"audio_{os.urandom(8).hex()}.wav")
 
-        # Generate MIDI
+        # Generate MIDI (reduced max_len for faster testing)
         if MODEL_LOADED:
-            generate_midi_with_model(prompt, midi_path, max_len=512, temperature=0.9)
+            generate_midi_with_model(prompt, midi_path, max_len=128, temperature=0.9)
         else:
             from midiutil import MIDIFile
             midi = MIDIFile(1)
@@ -156,14 +155,8 @@ def generate_music(prompt: str):
 
         # Convert to WAV
         if SOUNDFONT_PATH and midi_to_wav(midi_path, wav_path):
-            # Read WAV file and return as numpy array
-            sample_rate, audio_data = wavfile.read(wav_path)
-            # Convert to float32 for Gradio compatibility
-            if audio_data.dtype == np.int16:
-                audio_data = audio_data.astype(np.float32) / 32768.0
-            elif audio_data.dtype == np.int32:
-                audio_data = audio_data.astype(np.float32) / 2147483648.0
-            return (sample_rate, audio_data)
+            # Return file path - don't delete, let Gradio serve it
+            return wav_path
         else:
             return None
 
@@ -173,17 +166,12 @@ def generate_music(prompt: str):
         return None
 
     finally:
-        # Clean up temp files
+        # Only clean up MIDI file, keep WAV for Gradio to serve
         if midi_path and os.path.exists(midi_path):
             try:
                 os.unlink(midi_path)
             except:
                 pass
-        if wav_path and os.path.exists(wav_path):
-            try:
-                os.unlink(wav_path)
-            except:
-                pass
 
 # Create simple Gradio Interface
 demo = gr.Interface(
@@ -193,7 +181,7 @@ demo = gr.Interface(
         placeholder="A cheerful pop song with piano and drums in C major",
         lines=2
     ),
-    outputs=gr.Audio(label="Generated Music", type="numpy"),
+    outputs=gr.Audio(label="Generated Music", type="filepath"),
     title="VR Game Music Generator",
     description="Generate music from text descriptions using AI. Enter a prompt describing the music you want.",
     examples=[