L'électron rare
491e6ad8a4
## Chantier 1 — Voice Cloning - scripts/xtts_clone.py: Coqui XTTS-v2 voice cloning (6s sample → voix) - ws-chat.ts: synthesizeTTS avec détection sample → XTTS ou Piper - PersonaDetail.tsx: upload/delete sample vocal - app.ts: endpoints voice-sample (POST/GET/DELETE) ## Chantier 2 — Génération Musicale - scripts/compose_music.py: ACE-Step fallback MusicGen - /compose command (5min timeout, broadcast audio base64) - Chat.tsx: player <audio controls> inline - Eno persona (musique générative/ambient) - Pharmacius: routing @Eno Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
92 lines
3.2 KiB
Python
92 lines
3.2 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
KXKM_Clown — Music Generation via ACE-Step 1.5
|
|
|
|
Generates music from text prompts locally.
|
|
|
|
Usage:
|
|
python scripts/compose_music.py \
|
|
--prompt "ambient drone with deep bass, musique concrete style" \
|
|
--duration 30 \
|
|
--output /tmp/music.wav
|
|
|
|
Fallback to MusicGen if ACE-Step not installed.
|
|
Install: pip install ace-step OR pip install transformers scipy
|
|
"""
|
|
import argparse, json, os, sys, time
|
|
|
|
def parse_args():
|
|
p = argparse.ArgumentParser(description="KXKM Music Generation")
|
|
p.add_argument("--prompt", required=True)
|
|
p.add_argument("--duration", type=int, default=30, help="Duration in seconds")
|
|
p.add_argument("--output", required=True)
|
|
p.add_argument("--style", default="experimental", help="Style hint")
|
|
return p.parse_args()
|
|
|
|
def generate_with_musicgen(prompt, duration, output):
|
|
"""Fallback: use Meta's MusicGen (smaller, more available)."""
|
|
from transformers import AutoProcessor, MusicgenForConditionalGeneration
|
|
import scipy.io.wavfile
|
|
import numpy as np
|
|
|
|
print("[compose] Loading MusicGen small...", file=sys.stderr)
|
|
processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
|
|
model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
|
|
|
|
inputs = processor(text=[prompt], padding=True, return_tensors="pt")
|
|
# ~256 tokens per second of audio
|
|
max_tokens = min(duration * 256, 1536) # cap at ~6s for musicgen-small
|
|
|
|
print(f"[compose] Generating {max_tokens} tokens...", file=sys.stderr)
|
|
audio_values = model.generate(**inputs, max_new_tokens=max_tokens)
|
|
|
|
sampling_rate = model.config.audio_encoder.sampling_rate
|
|
audio_data = audio_values[0, 0].cpu().numpy()
|
|
audio_int16 = (audio_data * 32767).astype(np.int16)
|
|
|
|
scipy.io.wavfile.write(output, rate=sampling_rate, data=audio_int16)
|
|
return {"generator": "musicgen-small", "sampling_rate": sampling_rate}
|
|
|
|
def main():
|
|
args = parse_args()
|
|
start = time.time()
|
|
result = {"status": "failed", "error": None}
|
|
|
|
try:
|
|
os.makedirs(os.path.dirname(args.output) or ".", exist_ok=True)
|
|
|
|
full_prompt = f"{args.prompt}, {args.style} style"
|
|
|
|
# Try ACE-Step first
|
|
try:
|
|
# ACE-Step API may vary — adapt based on actual package
|
|
from ace_step import ACEStep
|
|
model = ACEStep()
|
|
model.generate(prompt=full_prompt, duration=args.duration, output_path=args.output)
|
|
gen_info = {"generator": "ace-step"}
|
|
except ImportError:
|
|
# Fallback to MusicGen
|
|
gen_info = generate_with_musicgen(full_prompt, args.duration, args.output)
|
|
|
|
duration = time.time() - start
|
|
file_size = os.path.getsize(args.output)
|
|
|
|
result = {
|
|
"status": "completed",
|
|
"outputFile": args.output,
|
|
"duration": round(duration, 2),
|
|
"fileSize": file_size,
|
|
"prompt": args.prompt[:200],
|
|
**gen_info,
|
|
}
|
|
print(f"[compose] Done in {duration:.1f}s -> {args.output} ({file_size} bytes)", file=sys.stderr)
|
|
|
|
except Exception as e:
|
|
result["error"] = str(e)
|
|
print(f"[compose] ERROR: {e}", file=sys.stderr)
|
|
|
|
print(json.dumps(result))
|
|
|
|
if __name__ == "__main__":
|
|
main()
|