import os import torch import torchaudio import psutil import time import sys import numpy as np import gc import gradio as gr from pydub import AudioSegment from audiocraft.models import MusicGen from torch.cuda.amp import autocast import warnings import random # Suppress warnings for cleaner output warnings.filterwarnings("ignore") # Set PYTORCH_CUDA_ALLOC_CONF to manage memory fragmentation os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128" # Check critical dependencies if np.__version__ != "1.23.5": print(f"WARNING: NumPy version {np.__version__} is being used. Tested with numpy==1.23.5.") if not torch.__version__.startswith(("2.1.0", "2.3.1")): print(f"WARNING: PyTorch version {torch.__version__} may not be compatible. Expected torch==2.1.0 or 2.3.1.") # 1) DEVICE SETUP device = "cuda" if torch.cuda.is_available() else "cpu" if device != "cuda": print("ERROR: CUDA is required for GPU rendering. CPU rendering is disabled.") sys.exit(1) print(f"CUDA is available. Using GPU: {torch.cuda.get_device_name(0)}") # Pre-run memory cleanup torch.cuda.empty_cache() gc.collect() torch.cuda.ipc_collect() torch.cuda.synchronize() # 2) LOAD MUSICGEN INTO VRAM try: print("Loading MusicGen medium model into VRAM...") local_model_path = "./models/musicgen-medium" if not os.path.exists(local_model_path): print(f"ERROR: Local model path {local_model_path} does not exist.") print("Please download the MusicGen medium model weights and place them in the correct directory.") sys.exit(1) musicgen_model = MusicGen.get_pretrained(local_model_path, device=device) musicgen_model.set_generation_params( duration=10, # Default chunk duration two_step_cfg=False # Disable two-step CFG for stability ) except Exception as e: print(f"ERROR: Failed to load MusicGen model: {e}") print("Ensure model weights are correctly placed and dependencies are installed.") sys.exit(1) # 3) RESOURCE MONITORING FUNCTION def print_resource_usage(stage: str): print(f"--- {stage} ---") print(f"GPU Memory Allocated: {torch.cuda.memory_allocated() / (1024**3):.2f} GB") print(f"GPU Memory Reserved: {torch.cuda.memory_reserved() / (1024**3):.2f} GB") print(f"CPU Memory Used: {psutil.virtual_memory().percent}%") print("---------------") # Check available GPU memory def check_vram_availability(required_gb=3.5): total_vram = torch.cuda.get_device_properties(0).total_memory / (1024**3) allocated_vram = torch.cuda.memory_allocated() / (1024**3) available_vram = total_vram - allocated_vram if available_vram < required_gb: print(f"WARNING: Low VRAM available ({available_vram:.2f} GB). Reduce total_duration or chunk_duration.") return available_vram >= required_gb # 4) GENRE PROMPT FUNCTIONS def set_red_hot_chili_peppers_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("strong rhythmic steps" if bpm > 120 else "groovy rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", groovy basslines" guitar = f", {guitar_style} guitar riffs" if guitar_style != "none" else ", syncopated guitar riffs" return f"Instrumental funk rock{bass}{guitar}{drum}{synth}, Red Hot Chili Peppers-inspired vibe with dynamic energy and funky breakdowns, {rhythm} at {bpm} BPM." def set_nirvana_grunge_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("intense rhythmic steps" if bpm > 120 else "grungy rhythmic pulse") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", melodic basslines" guitar = f", {guitar_style} guitar riffs" if guitar_style != "none" else ", raw distorted guitar riffs" return f"Instrumental grunge{bass}{guitar}{drum}{synth}, Nirvana-inspired angst-filled sound with quiet-loud dynamics, {rhythm} at {bpm} BPM." def set_pearl_jam_grunge_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("soulful rhythmic steps" if bpm > 120 else "driving rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", deep bass" guitar = f", {guitar_style} guitar leads" if guitar_style != "none" else ", soulful guitar leads" return f"Instrumental grunge{bass}{guitar}{drum}{synth}, Pearl Jam-inspired emotional intensity with soaring choruses, {rhythm} at {bpm} BPM." def set_soundgarden_grunge_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("heavy rhythmic steps" if bpm > 120 else "sludgy rhythmic groove") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else "" guitar = f", {guitar_style} guitar riffs" if guitar_style != "none" else ", heavy sludgy guitar riffs" return f"Instrumental grunge{bass}{guitar}{drum}{synth}, Soundgarden-inspired dark, psychedelic edge, {rhythm} at {bpm} BPM." def set_foo_fighters_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): styles = ["anthemic", "gritty", "melodic", "fast-paced", "driving"] tempos = ["upbeat", "mid-tempo", "high-energy"] moods = ["energetic", "introspective", "rebellious", "uplifting"] style = random.choice(styles) tempo = random.choice(tempos) mood = random.choice(moods) rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("powerful rhythmic steps" if bpm > 120 else "catchy rhythmic groove") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else "" guitar = f", {guitar_style} guitar riffs" if guitar_style != "none" else f", {style} guitar riffs" return f"Instrumental alternative rock{bass}{guitar}{drum}{synth}, Foo Fighters-inspired {mood} vibe with powerful choruses, {rhythm} at {bpm} BPM." def set_smashing_pumpkins_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("dynamic rhythmic steps" if bpm > 120 else "dreamy rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else "" guitar = f", {guitar_style} guitar textures" if guitar_style != "none" else ", dreamy guitar textures" return f"Instrumental alternative rock{bass}{guitar}{drum}{synth}, Smashing Pumpkins-inspired blend of melancholy and aggression, {rhythm} at {bpm} BPM." def set_radiohead_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("complex rhythmic steps" if bpm > 120 else "intricate rhythmic pulse") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else ", atmospheric synths" bass = f", {bass_style}" if bass_style != "none" else "" guitar = f", {guitar_style} guitar layers" if guitar_style != "none" else ", intricate guitar layers" return f"Instrumental experimental rock{bass}{guitar}{drum}{synth}, Radiohead-inspired blend of introspective and innovative soundscapes, {rhythm} at {bpm} BPM." def set_classic_rock_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("bluesy rhythmic steps" if bpm > 120 else "steady rhythmic groove") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", groovy bass" guitar = f", {guitar_style} electric guitars" if guitar_style != "none" else ", bluesy electric guitars" return f"Instrumental classic rock{bass}{guitar}{drum}{synth}, Led Zeppelin-inspired raw energy with dynamic solos, {rhythm} at {bpm} BPM." def set_alternative_rock_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("quirky rhythmic steps" if bpm > 120 else "energetic rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", melodic basslines" guitar = f", {guitar_style} guitar riffs" if guitar_style != "none" else ", distorted guitar riffs" return f"Instrumental alternative rock{bass}{guitar}{drum}{synth}, Pixies-inspired quirky, energetic vibe, {rhythm} at {bpm} BPM." def set_post_punk_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("sharp rhythmic steps" if bpm > 120 else "moody rhythmic pulse") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", driving basslines" guitar = f", {guitar_style} guitars" if guitar_style != "none" else ", jangly guitars" return f"Instrumental post-punk{bass}{guitar}{drum}{synth}, Joy Division-inspired moody, atmospheric sound, {rhythm} at {bpm} BPM." def set_indie_rock_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("catchy rhythmic steps" if bpm > 120 else "jangly rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else "" guitar = f", {guitar_style} guitars" if guitar_style != "none" else ", jangly guitars" return f"Instrumental indie rock{bass}{guitar}{drum}{synth}, Arctic Monkeys-inspired blend of catchy riffs, {rhythm} at {bpm} BPM." def set_funk_rock_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("aggressive rhythmic steps" if bpm > 120 else "funky rhythmic groove") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else "" bass = f", {bass_style}" if bass_style != "none" else ", slap bass" guitar = f", {guitar_style} guitar chords" if guitar_style != "none" else ", funky guitar chords" return f"Instrumental funk rock{bass}{guitar}{drum}{synth}, Rage Against the Machine-inspired mix of groove and aggression, {rhythm} at {bpm} BPM." def set_detroit_techno_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("pulsing rhythmic steps" if bpm > 120 else "deep rhythmic groove") drum = f", {drum_beat} drums" if drum_beat != "none" else ", crisp hi-hats" synth = f", {synthesizer} accents" if synthesizer != "none" else ", deep pulsing synths" bass = f", {bass_style}" if bass_style != "none" else ", driving basslines" guitar = f", {guitar_style} guitars" if guitar_style != "none" else "" return f"Instrumental Detroit techno{bass}{guitar}{drum}{synth}, Juan Atkins-inspired rhythmic groove, {rhythm} at {bpm} BPM." def set_deep_house_prompt(bpm, drum_beat, synthesizer, rhythmic_steps, bass_style, guitar_style): rhythm = f" with {rhythmic_steps}" if rhythmic_steps != "none" else ("soulful rhythmic steps" if bpm > 120 else "laid-back rhythmic flow") drum = f", {drum_beat} drums" if drum_beat != "none" else "" synth = f", {synthesizer} accents" if synthesizer != "none" else ", warm analog synth chords" bass = f", {bass_style}" if bass_style != "none" else ", deep basslines" guitar = f", {guitar_style} guitars" if guitar_style != "none" else "" return f"Instrumental deep house{bass}{guitar}{drum}{synth}, Larry Heard-inspired laid-back groove, {rhythm} at {bpm} BPM." # 5) AUDIO PROCESSING FUNCTIONS def apply_eq(segment): segment = segment.low_pass_filter(8000) segment = segment.high_pass_filter(80) return segment def apply_fade(segment, fade_in_duration=1000, fade_out_duration=1000): segment = segment.fade_in(fade_in_duration) segment = segment.fade_out(fade_out_duration) return segment # 6) GENERATION & I/O FUNCTIONS def generate_music(instrumental_prompt: str, cfg_scale: float, top_k: int, top_p: float, temperature: float, total_duration: int, chunk_duration: int, crossfade_duration: int, bpm: int, drum_beat: str, synthesizer: str, rhythmic_steps: str, bass_style: str, guitar_style: str): global musicgen_model if not instrumental_prompt.strip(): return None, "⚠️ Please enter a valid instrumental prompt!" try: start_time = time.time() total_duration = total_duration # Validated by radio button (30, 60, 90, 120) chunk_duration = min(max(chunk_duration, 5), 15) num_chunks = max(1, total_duration // chunk_duration) chunk_duration = total_duration / num_chunks overlap_duration = min(1.0, crossfade_duration / 1000.0) generation_duration = chunk_duration + overlap_duration sample_rate = musicgen_model.sample_rate audio_segments = [] if not check_vram_availability(required_gb=3.5): return None, "⚠️ Insufficient VRAM for generation. Reduce total_duration or chunk_duration." print("Generating audio...") seed = 42 torch.manual_seed(seed) np.random.seed(seed) for i in range(num_chunks): chunk_prompt = instrumental_prompt print(f"Generating chunk {i+1}/{num_chunks} on GPU (prompt: {chunk_prompt})...") musicgen_model.set_generation_params( duration=generation_duration, use_sampling=True, top_k=top_k, top_p=top_p, temperature=temperature, cfg_coef=cfg_scale ) print_resource_usage(f"Before Chunk {i+1} Generation") with torch.no_grad(): with autocast(): audio_chunk = musicgen_model.generate([chunk_prompt], progress=True)[0] audio_chunk = audio_chunk.cpu().to(dtype=torch.float32) if audio_chunk.dim() == 1: audio_chunk = torch.stack([audio_chunk, audio_chunk], dim=0) elif audio_chunk.dim() == 2 and audio_chunk.shape[0] == 1: audio_chunk = torch.cat([audio_chunk, audio_chunk], dim=0) elif audio_chunk.dim() == 2 and audio_chunk.shape[0] != 2: audio_chunk = audio_chunk[:1, :] audio_chunk = torch.cat([audio_chunk, audio_chunk], dim=0) elif audio_chunk.dim() > 2: audio_chunk = audio_chunk.view(2, -1) if audio_chunk.shape[0] != 2: raise ValueError(f"Expected stereo audio with shape (2, samples), got shape {audio_chunk.shape}") temp_wav_path = f"temp_chunk_{i}.wav" torchaudio.save(temp_wav_path, audio_chunk, sample_rate, bits_per_sample=24) segment = AudioSegment.from_wav(temp_wav_path) os.remove(temp_wav_path) audio_segments.append(segment) torch.cuda.empty_cache() gc.collect() torch.cuda.ipc_collect() torch.cuda.synchronize() time.sleep(0.5) print_resource_usage(f"After Chunk {i+1} Generation") print("Combining audio chunks...") final_segment = audio_segments[0] for i in range(1, len(audio_segments)): next_segment = audio_segments[i] next_segment = next_segment + 1 final_segment = final_segment.append(next_segment, crossfade=crossfade_duration) final_segment = final_segment[:total_duration * 1000] print("Post-processing final track...") final_segment = apply_eq(final_segment) final_segment = final_segment.normalize(headroom=-9.0) final_segment = apply_fade(final_segment) mp3_path = "output_cleaned.mp3" final_segment.export( mp3_path, format="mp3", bitrate="128k", tags={"title": "GhostAI Instrumental", "artist": "GhostAI"} ) print(f"Saved final audio to {mp3_path}") print_resource_usage("After Final Generation") print(f"Total Generation Time: {time.time() - start_time:.2f} seconds") return mp3_path, "✅ Done! Generated instrumental audio." except Exception as e: return None, f"❌ Generation failed: {e}" finally: torch.cuda.empty_cache() gc.collect() torch.cuda.ipc_collect() torch.cuda.synchronize() # Function to clear inputs def clear_inputs(): return "", 3.0, 250, 0.9, 1.0, 30, 10, 1000, 120, "none", "none", "none", "none", "none" # 7) CUSTOM CSS css = """ body { background: linear-gradient(135deg, #0A0A0A 0%, #1C2526 100%); color: #E0E0E0; font-family: 'Orbitron', sans-serif; } .header-container { text-align: center; padding: 10px 20px; background: rgba(0, 0, 0, 0.9); border-bottom: 1px solid #00FF9F; } #ghost-logo { font-size: 40px; animation: glitch-ghost 1.5s infinite; } h1 { color: #A100FF; font-size: 24px; animation: glitch-text 2s infinite; } p { color: #E0E0E0; font-size: 12px; } .input-container, .settings-container, .output-container { max-width: 1200px; margin: 20px auto; padding: 20px; background: rgba(28, 37, 38, 0.8); border-radius: 10px; } .textbox { background: #1A1A1A; border: 1px solid #A100FF; color: #E0E0E0; } .genre-buttons { display: flex; justify-content: center; flex-wrap: wrap; gap: 15px; } .genre-btn, button { background: linear-gradient(45deg, #A100FF, #00FF9F); border: none; color: #0A0A0A; padding: 10px 20px; border-radius: 5px; } .gradio-container { padding: 20px; } .group-container { margin-bottom: 20px; padding: 15px; border: 1px solid #00FF9F; border-radius: 8px; } @keyframes glitch-ghost { 0% { transform: translate(0, 0); opacity: 1; } 20% { transform: translate(-5px, 2px); opacity: 0.8; } 100% { transform: translate(0, 0); opacity: 1; } } @keyframes glitch-text { 0% { transform: translate(0, 0); } 20% { transform: translate(-2px, 1px); } 100% { transform: translate(0, 0); } } @font-face { font-family: 'Orbitron'; src: url('https://fonts.gstatic.com/s/orbitron/v29/yMJRMIlzdpvBhQQL_Qq7dy0.woff2') format('woff2'); } """ # 8) BUILD WITH BLOCKS with gr.Blocks(css=css) as demo: gr.Markdown("""
Summon the Sound of the Unknown