Datasets:
Tasks:
Audio Classification
Formats:
parquet
Size:
1K - 10K
ArXiv:
Tags:
arxiv:2606.01686
music
ai-generated-music
ai-generated-music-detection
plagiarism-detection
ace-step
License:
File size: 11,073 Bytes
b347b70 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 | #!/usr/bin/env python3
"""Fill missing Lyria-3 tracks by retrying failed indices with backoff."""
import json, os, os, time, random
from pathlib import Path
from google import genai
from google.genai import types
API_KEY = os.environ["GOOGLE_API_KEY"] # set your own key
OUT_DIR = Path("/ssd_data/dataset/haim_dataset/A_full_generation/A2_fake/lyria-pro3")
TARGET = 6000
MAX_RETRIES = 5
PROMPTS = [
"An epic cinematic orchestral piece about a journey home. Starts with a solo piano intro, builds through sweeping strings, and climaxes with a massive wall of sound.",
"A dreamy indie pop song about city nights. Soft synths and reverb-heavy guitars in the verse, building to a bright, anthemic chorus with layered vocals.",
"A soulful R&B ballad about lost love. Warm Rhodes piano and gentle finger-picked guitar in the verse, swelling strings and harmonies in the chorus.",
"An energetic K-pop dance track with punchy synth bass, crisp electronic drums, and catchy vocal hooks. Features a rap bridge and a high-energy final chorus.",
"A melancholic folk song with fingerpicked acoustic guitar and soft harmonica. The singer tells a story of leaving a small town, with a bittersweet chorus.",
"A high-energy punk rock anthem with distorted power chords, fast drums, and shouted vocals about rebellion and freedom.",
"A smooth jazz vocal piece with walking bass, brushed drums, and a sultry saxophone solo between verses.",
"A gospel choir performance with powerful organ chords, hand claps, and call-and-response vocals building to a triumphant climax.",
"A reggae song with deep dub bass, offbeat guitar skanks, and laid-back vocals about sunshine and good vibes. Features an echo-drenched melodica solo.",
"A country western ballad with steel guitar, acoustic strumming, and heartfelt lyrics about a long drive through open plains.",
"An atmospheric ambient instrumental piece. Evolving synthesizer pads, distant piano notes, and subtle field recordings create a meditative soundscape. Instrumental only, no vocals.",
"A funky instrumental groove with slap bass, wah-wah guitar, and tight horn stabs. Think 1970s funk band jamming. Instrumental only, no vocals.",
"A classical string quartet performing a dramatic piece in D minor. Starts adagio with a melancholic cello melody, builds to an allegro with all four instruments in passionate dialogue. Instrumental only, no vocals.",
"A lo-fi hip hop instrumental beat with dusty vinyl crackle, mellow Rhodes piano chords, and a lazy boom-bap drum pattern. Instrumental only, no vocals.",
"An intense drum and bass track with rapid breakbeats, deep sub bass, and glitchy synth textures. Instrumental only, no vocals.",
"A flamenco guitar solo piece with passionate rasgueado strumming, percussive golpe taps, and intricate picado runs. Instrumental only, no vocals.",
"A minimalist techno track with hypnotic repetitive synth patterns, deep kick drum, and slowly evolving filter sweeps. Instrumental only, no vocals.",
"A baroque harpsichord piece in the style of Bach, featuring intricate counterpoint and ornamental trills. Instrumental only, no vocals.",
"An afrobeat instrumental jam with polyrhythmic percussion, driving bass guitar, and bright horn arrangements. Instrumental only, no vocals.",
"A chiptune 8-bit video game soundtrack with energetic square wave melodies and arpeggiated basslines. Instrumental only, no vocals.",
"""[0:00 - 0:15] Intro: Soft acoustic guitar arpeggio with ambient reverb.
[0:15 - 0:45] Verse 1: A male voice sings about walking through autumn leaves, accompanied by gentle piano and light percussion.
[0:45 - 1:15] Chorus: Full band enters with drums, bass, and electric guitar. The melody soars with hope and nostalgia.
[1:15 - 1:45] Verse 2: The texture thins back to guitar and voice, with added cello.
[1:45 - 2:15] Chorus: Bigger and more emotional, with backing vocal harmonies.
[2:15 - 2:30] Outro: Fade out with solo piano.""",
"""[0:00 - 0:10] Intro: Begin with a pulsing electronic bass and hi-hat pattern.
[0:10 - 0:40] Verse: Dark, moody synth pads with a whispered vocal about neon-lit streets.
[0:40 - 1:10] Chorus: Explosive drop with heavy bass, layered synths, and powerful vocals.
[1:10 - 1:40] Bridge: Stripped back to just voice and a single synth pad, building tension.
[1:40 - 2:10] Final Chorus: Even bigger than before with added vocal harmonies and orchestral hits.
[2:10 - 2:20] Outro: The bass pulse fades into silence.""",
"A romantic French chanson about a Parisian evening. Accordion, nylon guitar, and a warm baritone voice singing about love along the Seine river.",
"A passionate Spanish flamenco pop song with acoustic guitar, cajon percussion, and a powerful female voice singing about heartbreak and resilience.",
"A Japanese city pop track with groovy bass, bright electric piano, and smooth vocals. The style is reminiscent of 1980s Tokyo nightlife music.",
"A Brazilian bossa nova piece with gentle nylon guitar, soft brushed drums, and a breathy vocal singing in Portuguese about a tropical sunset.",
"A Korean ballad with emotional piano, sweeping string arrangement, and a powerful vocal performance building from intimate verse to soaring chorus.",
"A Bollywood dance number with dhol drums, synthesizers, and energetic Hindi vocals. Features a catchy hook and a rhythmic breakdown.",
"A traditional Irish folk song with tin whistle, bodhran drum, and a fiddle. A pub choir joins in the chorus with hearty harmonies.",
"An Arabic song with oud, qanun, and darbuka percussion. A rich tenor voice sings an ornate melody with traditional maqam scales.",
"A progressive rock epic that shifts between 7/8 and 4/4 time signatures. Features a Mellotron intro, complex guitar harmonies, a Hammond organ solo, and a thunderous drum fill leading into the final section.",
"A trap beat with booming 808 bass, skittering hi-hat triplets, dark piano chords, and an Auto-Tuned vocal hook about late nights and ambition.",
"A synthwave retro track with warm analog synth pads, gated reverb drums, and a driving arpeggiated bassline. The mood is nostalgic and cinematic, like driving through a neon city at night. Instrumental only, no vocals.",
"A grunge rock song with heavily distorted guitars, a loose and powerful drum groove, and raw, anguished vocals about disillusionment. Quiet verse, loud chorus dynamic.",
"A deep house track with a four-on-the-floor kick, warm filtered chords, a rolling bassline, and chopped vocal samples creating a hypnotic groove. Instrumental only, no vocals.",
"A neo-soul piece with a jazzy chord progression on Wurlitzer electric piano, a pocket groove from bass and drums, and silky vocals with subtle runs and ad-libs.",
"A psychedelic rock jam with wah-wah guitar, phaser effects, droning sitar, and swirling Leslie speaker organ. The song builds from a slow groove into a wild improvised climax. Instrumental only, no vocals.",
"A new age meditation piece with Tibetan singing bowls, gentle synthesizer drones, and the sound of flowing water. Peaceful and transcendent. Instrumental only, no vocals.",
"An EDM festival anthem starting with a filtered build-up, clapping rhythm, and pitched vocal chops, exploding into a massive supersaw drop with sidechained bass.",
"A bluegrass jam with rapid banjo picking, fiddle, mandolin, and upright bass. The tempo is fast and the energy is infectious, with each instrument taking a solo break.",
]
def main():
client = genai.Client(api_key=API_KEY)
existing = {int(f.stem.split("_")[1]) for f in OUT_DIR.glob("lyria3_*.mp3")}
missing = sorted(set(range(TARGET)) - existing)
print(f"Missing: {len(missing)} tracks")
if not missing:
print("All 6000 tracks present!")
return
meta_path = OUT_DIR / "metadata_fill.jsonl"
meta_f = open(meta_path, "a", encoding="utf-8")
import threading
from concurrent.futures import ThreadPoolExecutor, as_completed
lock = threading.Lock()
filled = 0
def generate_one(i):
prompt = PROMPTS[i % len(PROMPTS)]
fname = f"lyria3_{i:05d}.mp3"
out_path = OUT_DIR / fname
for attempt in range(MAX_RETRIES):
try:
response = client.models.generate_content(
model="lyria-3-pro-preview",
contents=prompt,
config=types.GenerateContentConfig(
response_modalities=["AUDIO", "TEXT"],
),
)
lyrics = []
audio_data = None
if response.parts:
for part in response.parts:
if part.text is not None:
lyrics.append(part.text)
elif part.inline_data is not None:
audio_data = part.inline_data.data
if audio_data:
with open(out_path, "wb") as f:
f.write(audio_data)
meta = {
"track_id": f"lyria3_{i:05d}",
"filename": fname,
"platform": "google_lyria3",
"model_name": "Lyria-3 Pro",
"model_version": "pro-preview",
"prompt": prompt[:200],
"lyrics": "\n".join(lyrics) if lyrics else None,
"audio_size_bytes": len(audio_data),
"collection_method": "generate",
"generated_at": time.strftime("%Y-%m-%dT%H:%M:%S"),
}
with open(OUT_DIR / f"lyria3_{i:05d}.json", "w", encoding="utf-8") as jf:
json.dump(meta, jf, ensure_ascii=False, indent=2)
with lock:
meta_f.write(json.dumps(meta, ensure_ascii=False) + "\n")
meta_f.flush()
return "ok"
# Empty response, retry with backoff
time.sleep(2 ** attempt)
except Exception as e:
time.sleep(2 ** attempt)
if attempt == MAX_RETRIES - 1:
return f"err:{e}"
return "empty_after_retries"
WORKERS = 5
with ThreadPoolExecutor(max_workers=WORKERS) as pool:
futures = {pool.submit(generate_one, i): i for i in missing}
for fut in as_completed(futures):
result = fut.result()
filled += 1
if filled % 20 == 0 or filled == len(missing):
mp3s = len(list(OUT_DIR.glob("*.mp3")))
print(f"[{mp3s}/{TARGET}] filled={filled}/{len(missing)} last={result}")
meta_f.close()
final = len(list(OUT_DIR.glob("*.mp3")))
print(f"Done: {final}/{TARGET} tracks")
if final < TARGET:
print(f"Still missing {TARGET - final}. Run again to retry.")
if __name__ == "__main__":
main()
|