Update app.py

app.py

@@ -1,90 +1,54 @@
-import torch
-from torch.serialization import add_safe_globals
-import numpy.core.multiarray
-
-# Patch for PyTorch 2.6 to allow loading Bark's model checkpoints
-add_safe_globals({
-    'scalar': numpy.core.multiarray.scalar
-})
-
-
-import os
-import shutil
-import tempfile
 import streamlit as st
 import torchaudio
-from pydub import AudioSegment
-from bark import generate_audio, SAMPLE_RATE as BARK_RATE
+import torch
 from audiocraft.models import MusicGen
+from pydub import AudioSegment
+import os
 
-# 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')
-
-# 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("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 = 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  # use cloned vocals
-
-    # 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)
-        mixed = instrumental_seg.overlay(vocals_seg.set_frame_rate(32000).set_channels(1))
-        final_output_path = tempfile.mktemp(suffix=".wav")
-        mixed.export(final_output_path, format="wav")
-
-    st.audio(final_output_path)
-    st.success("Your song is ready!")
+st.set_page_config(page_title="Suno-like AI Music Generator", layout="centered")
+st.title("Suno-like AI Music Generator")
+
+# Load the pre-recorded male voice sample
+st.subheader("Step 1: Pre-recorded AI Voice Sample")
+voice_path = "sample_voice.mp3"
+
+if os.path.exists(voice_path):
+    audio_file = open(voice_path, 'rb')
+    audio_bytes = audio_file.read()
+    st.audio(audio_bytes, format='audio/mp3')
+else:
+    st.error("Voice sample not found. Please make sure 'sample_voice.mp3' is in the same directory.")
+
+# Generate Instrumental with MusicGen
+st.subheader("Step 2: Generate Instrumental with MusicGen")
+prompt = st.text_input("Enter music description (e.g. energetic trap beat with dark mood)", "emotional trap beat with dark pads and heavy drums")
+
+if st.button("Generate Music"):
+    with st.spinner("Generating music..."):
+        model = MusicGen.get_pretrained("facebook/musicgen-small")
+        model.set_generation_params(duration=10)
+        output = model.generate([prompt])
+        output_path = "musicgen_output.wav"
+        torchaudio.save(output_path, output[0].cpu(), 32000)
+        st.audio(output_path, format="audio/wav")
+        st.success("Music generation complete!")
+
+# Combine voice and beat (optional step to simulate Suno-like output)
+st.subheader("Step 3: Simulated Combination")
+if os.path.exists("musicgen_output.wav") and os.path.exists("sample_voice.mp3"):
+    if st.button("Combine Voice + Instrumental"):
+        # Convert MP3 voice to WAV for merging
+        voice = AudioSegment.from_mp3("sample_voice.mp3")
+        beat = AudioSegment.from_wav("musicgen_output.wav")
+
+        # Adjust volumes and overlay
+        voice = voice - 4
+        beat = beat - 2
+        combined = beat.overlay(voice, loop=False)
+        combined_path = "combined_output.mp3"
+        combined.export(combined_path, format="mp3")
+
+        st.audio(combined_path, format="audio/mp3")
+        st.success("Combined output ready!")
+else:
+    st.info("Generate the beat and make sure voice sample is available.")