Update app.py

app.py

@@ -1,44 +1,74 @@
+import os
+import shutil
+import tempfile
 import streamlit as st
 import torchaudio
-import tempfile
-import os
 from pydub import AudioSegment
 from bark import generate_audio, SAMPLE_RATE as BARK_RATE
 from audiocraft.models import MusicGen
 
-# Load MusicGen model
+# Use smaller Bark models for better compatibility
+os.environ["SUNO_USE_SMALL_MODELS"] = "1"
+BARK_CACHE = os.path.expanduser("~/.cache/suno/")
+
+# Load MusicGen model once
 musicgen = MusicGen.get_pretrained('facebook/musicgen-small')
 
-st.set_page_config(page_title="Suno Clone", layout="centered")
-st.title("Suno-like Music Generator")
-st.markdown("Combining **Bark TTS**, **RVC Voice Cloning**, and **MusicLM** for full track creation.")
+# Streamlit layout
+st.set_page_config(page_title="Suno-Like Generator", layout="centered")
+st.title("Suno-Like Music Generator")
+st.markdown("Powered by **Bark + RVC + MusicGen**")
 
+# Inputs
 lyrics = st.text_area("Enter your lyrics:", height=150)
-genre_prompt = st.text_input("Enter Music Style (e.g., 'trap beat with piano')", value="trap beat with piano")
-uploaded_voice = st.file_uploader("Upload your voice sample (for cloning - WAV only)", type=["wav"])
+genre_prompt = st.text_input("Describe the music style (e.g., 'afrobeats with guitar')", value="trap beat with piano")
+uploaded_voice = st.file_uploader("Upload your voice sample for cloning (WAV)", type=["wav"])
+
+# Optional dev button
+if st.sidebar.button("Clear Bark Cache"):
+    shutil.rmtree(BARK_CACHE, ignore_errors=True)
+    st.success("Bark model cache cleared. Reload app to redownload models.")
 
+# Bark fallback-safe function
+def safe_bark_generate(text, speaker="en_speaker_6"):
+    try:
+        return generate_audio(text, history_prompt=speaker)
+    except Exception as e:
+        st.warning(f"Speaker {speaker} failed, retrying with default. Error: {e}")
+        try:
+            return generate_audio(text)
+        except Exception as e2:
+            st.error(f"Bark generation failed. Error: {e2}")
+            return None
+
+# Main generation block
 if st.button("Generate Song"):
+    # Bark TTS
     with st.spinner("Generating vocals with Bark..."):
-        vocals_tensor = generate_audio(lyrics, history_prompt="v2/en_speaker_6")
+        vocals_tensor = safe_bark_generate(lyrics)
+        if vocals_tensor is None:
+            st.stop()
         vocals_path = tempfile.mktemp(suffix=".wav")
         torchaudio.save(vocals_path, vocals_tensor.squeeze(0).cpu(), BARK_RATE)
 
+    # RVC Voice Cloning
     if uploaded_voice:
         with st.spinner("Cloning voice with RVC..."):
             uploaded_voice_path = tempfile.mktemp(suffix=".wav")
             with open(uploaded_voice_path, "wb") as f:
                 f.write(uploaded_voice.read())
-
             cloned_path = tempfile.mktemp(suffix=".wav")
             os.system(f"python infer_rvc.py --input \"{vocals_path}\" --voice \"{uploaded_voice_path}\" --output \"{cloned_path}\"")
-            vocals_path = cloned_path
+            vocals_path = cloned_path  # use cloned vocals
 
-    with st.spinner("Generating instrumental with MusicLM..."):
+    # MusicGen Instrumental
+    with st.spinner("Generating instrumental with MusicGen..."):
         musicgen.set_generation_params(duration=15)
         music = musicgen.generate([genre_prompt])
         instrumental_path = tempfile.mktemp(suffix=".wav")
         torchaudio.save(instrumental_path, music[0].cpu(), 32000)
 
+    # Mix vocals + instrumental
     with st.spinner("Mixing vocals and instrumental..."):
         vocals_seg = AudioSegment.from_wav(vocals_path)
         instrumental_seg = AudioSegment.from_wav(instrumental_path)