Spaces:
Running
Running
File size: 21,589 Bytes
45d92cc efe8555 45d92cc 4d426ef 45d92cc a07ee41 45d92cc efe8555 45d92cc efe8555 45d92cc 4d426ef 45d92cc a07ee41 45d92cc efe8555 a07ee41 efe8555 45d92cc 518c021 45d92cc efe8555 45d92cc efe8555 45d92cc 0a82bbc 45d92cc 4d426ef 45d92cc 0a82bbc 45d92cc 4d426ef a07ee41 4d426ef 45d92cc | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 | from __future__ import annotations
import json
import os
import re
import urllib.error
import urllib.request
from pathlib import Path
from typing import Any, Literal, Protocol
from huggingface_hub import InferenceClient
from pydantic import BaseModel, Field
from studio.adapters.router import route_id_for_model
from studio.registry import ModelRegistry, load_registry
PROMPT_PATH = Path(__file__).resolve().parents[1] / "prompts" / "producer.md"
MASTER_PROMPT_PATH = Path(__file__).resolve().parents[1] / "prompts" / "mastering.md"
_FALLBACK_PRODUCER = (
"Du bist der Musikproducer-Agent von 2B-Studio. Antworte ausschließlich als JSON "
"für fill_create_fields. Spiegele die Sprache des Users. Keine freien Generierungen starten."
)
_FALLBACK_MASTER = (
"Du bist der Mastering-Agent von 2B-Studio. Antworte ausschließlich als JSON "
"für fill_master_fields. Spiegele die Sprache des Users. Kein Mastering selbst starten."
)
def _read_prompt(env_key: str, path: Path, fallback: str) -> str:
env = (os.environ.get(env_key) or "").strip()
if env and not env.startswith("# STUB"):
return env
if path.is_file():
text = path.read_text(encoding="utf-8").strip()
if text and not text.startswith("# STUB"):
return text
return fallback
def gemini_api_key() -> str | None:
return os.environ.get("GEMINI_API_KEY", "").strip() or None
class FieldProposal(BaseModel):
model_id: str | None = None
genre_style_instruments: str
lyrics: str
negative: str
rationale: str
duration_s: int = Field(default=20, ge=5, le=180)
idea: str = ""
studio: dict[str, str] = Field(default_factory=dict)
renderings: dict[str, str] = Field(default_factory=dict)
preview: str = ""
ask_user: str | None = None
warnings: list[str] = Field(default_factory=list)
source: Literal["gemini", "hf-inference", "fallback"] = "fallback"
class ProducerAction(BaseModel):
type: Literal["fill_create_fields"] = "fill_create_fields"
payload: FieldProposal
preview: str
ask_user: str | None = None
class MasterProposal(BaseModel):
profile: Literal[
"streaming_safe",
"streaming_loud",
"club",
"dynamic_hifi",
"premium_clarity",
"audiophile_preserve",
] = "streaming_safe"
purity_mode: Literal["soft_keep_air", "strict", "off"] = "soft_keep_air"
keep_air_db: float = Field(default=-20.0, ge=-48, le=0)
margin: float = Field(default=2.0, ge=0.5, le=2.9)
intensity: int = Field(default=50, ge=0, le=100)
analyze_only: bool = False
rationale: str
preview: str
requires_strict_approval: bool = False
warnings: list[str] = Field(default_factory=list)
class MasterAction(BaseModel):
type: Literal["fill_master_params"] = "fill_master_params"
payload: MasterProposal
preview: str
class ProducerProvider(Protocol):
def propose(self, message: str, current_fields: dict[str, Any]) -> dict[str, Any]: ...
class HFInferenceProducerProvider:
def __init__(self) -> None:
token = os.environ.get("HF_TOKEN", "")
if not token:
raise ValueError("HF_TOKEN ist nicht gesetzt.")
self.model_id = os.environ.get(
"AGENT_MODEL_ID", "meta-llama/Llama-3.2-3B-Instruct"
)
self.client = InferenceClient(token=token)
self.system_prompt = _read_prompt("PRODUCER_SYSTEM_PROMPT", PROMPT_PATH, _FALLBACK_PRODUCER)
def propose(self, message: str, current_fields: dict[str, Any]) -> dict[str, Any]:
response = self.client.chat.completions.create(
model=self.model_id,
messages=[
{"role": "system", "content": self.system_prompt},
{
"role": "user",
"content": json.dumps(
{"idea": message, "current_fields": current_fields},
ensure_ascii=False,
),
},
],
temperature=0.2,
max_tokens=1400,
)
content = response.choices[0].message.content or "{}"
cleaned = re.sub(r"^```(?:json)?\s*|\s*```$", "", content.strip())
parsed = json.loads(cleaned)
if not isinstance(parsed, dict):
raise ValueError("Der Agent lieferte kein JSON-Objekt.")
payload = parsed.get("payload", parsed)
if not isinstance(payload, dict):
raise ValueError("Der Agent lieferte kein gültiges Payload-Objekt.")
return payload
class GeminiProducerProvider:
"""Gemini via REST API (keine Zusatz-Dependency, Free-Tier-tauglich)."""
def __init__(self) -> None:
key = gemini_api_key()
if not key:
raise ValueError("GEMINI_API_KEY ist nicht gesetzt.")
self.api_key = key
self.model_id = os.environ.get("GEMINI_MODEL", "gemini-3.5-flash")
self.system_prompt = _read_prompt("PRODUCER_SYSTEM_PROMPT", PROMPT_PATH, _FALLBACK_PRODUCER)
self.base_url = "https://generativelanguage.googleapis.com/v1beta"
def propose(self, message: str, current_fields: dict[str, Any]) -> dict[str, Any]:
url = (
f"{self.base_url}/models/{self.model_id}:generateContent"
f"?key={self.api_key}"
)
body = {
"system_instruction": {"parts": [{"text": self.system_prompt}]},
"contents": [
{
"parts": [
{
"text": json.dumps(
{"idea": message, "current_fields": current_fields},
ensure_ascii=False,
)
}
]
}
],
"generationConfig": {
"temperature": 0.2,
"maxOutputTokens": 4096,
"responseMimeType": "application/json",
},
}
request = urllib.request.Request(
url,
data=json.dumps(body).encode("utf-8"),
headers={"Content-Type": "application/json"},
method="POST",
)
try:
with urllib.request.urlopen(request, timeout=60) as response:
payload = json.loads(response.read().decode("utf-8"))
except urllib.error.HTTPError as error:
detail = error.read().decode("utf-8", errors="replace")[:300]
raise ValueError(f"Gemini HTTP {error.code}: {detail}") from error
except urllib.error.URLError as error:
raise ValueError(f"Gemini nicht erreichbar: {error.reason}") from error
candidates = payload.get("candidates") or []
if not candidates:
block = payload.get("promptFeedback", {}).get("blockReason")
raise ValueError(f"Gemini lieferte keine Antwort (blockReason={block}).")
text = candidates[0].get("content", {}).get("parts", [{}])[0].get("text", "")
cleaned = re.sub(r"^```(?:json)?\s*|\s*```$", "", text.strip())
parsed = json.loads(cleaned)
if not isinstance(parsed, dict):
raise ValueError("Gemini lieferte kein JSON-Objekt.")
payload_out = parsed.get("payload", parsed)
if not isinstance(payload_out, dict):
raise ValueError("Gemini lieferte kein gültiges Payload-Objekt.")
return payload_out
def _duration_from_message(message: str) -> int:
match = re.search(
r"\b(?P<value>\d{1,3})\s*(?P<unit>sekunden?|seconds?|secs?|s|minuten?|minutes?|min)\b",
message.casefold(),
)
if not match:
return 20
value = int(match.group("value"))
if match.group("unit").startswith(("min", "minute")):
value *= 60
return max(5, min(180, value))
def _fallback_proposal(message: str) -> FieldProposal:
normalized = message.casefold()
duration_s = _duration_from_message(message)
vocals_requested = any(
term in normalized
for term in (
"vocal",
"gesang",
"sänger",
"sängerin",
"männlich",
"weiblich",
"lyrics",
"stimme",
)
) and "instrumental" not in normalized and not any(
neg in normalized
for neg in (
"ohne gesang", "kein gesang", "keinen gesang", "ohne vocals", "keine vocals", "ohne stimme",
"ohne text", "keine lyrics", "ohne lyrics", "no vocals", "without vocals", "no lyrics",
)
)
if any(term in normalized for term in ("riddim", "reggae", "dub")):
style = (
"dark roots riddim, 140 BPM, analog sub bass, tight live drums, "
"warm organ skanks, evolving dub transitions"
)
elif any(term in normalized for term in ("techno", "club", "rave")):
style = (
"hypnotic techno, 132 BPM, rolling sub bass, crisp percussion, "
"warehouse depth, tension-building arrangement"
)
elif any(term in normalized for term in ("lofi", "chill", "entspannt")):
style = (
"lo-fi soul, 82 BPM, dusty drums, mellow Rhodes piano, warm tape texture, "
"human timing, intimate dynamics"
)
elif any(term in normalized for term in ("film", "cinematic", "episch")):
style = (
"cinematic electronica, 96 BPM, evolving strings, deep percussion, "
"wide synth layers, clear emotional arc"
)
elif any(term in normalized for term in ("rock", "gitarre", "guitar")):
style = (
"modern alternative rock, 118 BPM, expressive electric guitars, live drums, "
"driving bass, dynamic verse-to-chorus lift"
)
else:
style = (
"modern electronic soul, 108 BPM, punchy drums, warm bass, memorable hook, "
"intentional transitions, polished stereo depth"
)
if vocals_requested:
style += ", expressive lead vocal with clear diction"
lyrics = (
"[Verse]\nWe turn the quiet into light\nEvery step is drawing near\n\n"
"[Chorus]\nWe hold the line tonight\nMake the hidden signal clear"
)
model_id = "minimax-music3"
else:
lyrics = "[Instrumental]"
model_id = "2b-musicgen-medium" if duration_s > 20 else "2b-musicgen-small"
short_caption = style[:800]
return FieldProposal(
model_id=model_id,
genre_style_instruments=style,
lyrics=lyrics,
negative=(
"muddy low end, harsh sibilance, uncontrolled clipping, phase cancellation, "
"static arrangement, generic AI artifacts"
),
duration_s=duration_s,
idea=message.strip(),
studio={
"global_metadata": f"Basic Attributes: {style}",
"vocal_details": (
"Vocal Gender & Timbre: expressive lead, clear diction"
if vocals_requested
else "Vocal Direction: instrumental, no lead vocals"
),
"arrangement": "Instrument Lifecycle: focused opening, developed middle, resolved ending",
},
renderings={
"musicgen_prompt": style,
"short_caption": short_caption,
},
preview=f"{duration_s}s · {'Vocal' if vocals_requested else 'Instrumental'} · {model_id}",
rationale=(
"MiniMax passt zur strukturierten Vocal-Anforderung."
if vocals_requested
else "Der live angebundene 2B-Audio-Worker passt zum instrumentalen Entwurf."
),
source="fallback",
)
class ProducerAgent:
def __init__(
self,
registry: ModelRegistry,
provider: ProducerProvider | None = None,
) -> None:
self.registry = registry
self.provider = provider
if self.provider is None:
if gemini_api_key():
self.provider = GeminiProducerProvider()
elif os.environ.get("HF_TOKEN"):
self.provider = HFInferenceProducerProvider()
def propose(
self,
message: str,
current_fields: dict[str, Any] | None = None,
) -> ProducerAction:
warnings: list[str] = []
if self.provider is not None:
try:
raw = self.provider.propose(message, current_fields or {})
source = (
"gemini"
if isinstance(self.provider, GeminiProducerProvider)
else "hf-inference"
)
proposal = FieldProposal.model_validate({**raw, "source": source})
except Exception as error:
proposal = _fallback_proposal(message)
warnings.append(f"LLM-Fallback aktiv: {type(error).__name__}")
else:
proposal = _fallback_proposal(message)
# Deterministische Absicherung: vage Eingabe -> immer eine Rückfrage
if proposal.ask_user is None and _is_vague_input(message):
proposal.ask_user = _default_ask(message)
try:
model = self.registry.get(proposal.model_id or "")
except KeyError:
model = self.registry.get("2b-musicgen-small")
proposal.model_id = model.id
warnings.append("Unbekanntes Agentenmodell wurde auf 2B Audio Small zurückgesetzt.")
if proposal.duration_s > model.max_duration_s:
warnings.append(
f"Dauer wurde an {model.label} angepasst: maximal {model.max_duration_s} s."
)
proposal.duration_s = model.max_duration_s
if proposal.lyrics.strip().casefold() != "[instrumental]" and not model.supports_vocals:
warnings.append(f"{model.label} unterstützt Gesang nur eingeschränkt.")
if not route_id_for_model(model).endswith("-real"):
warnings.append(
f"{model.label} ist gerade nicht erreichbar; der Worker antwortet nicht."
)
proposal.warnings = [*proposal.warnings, *warnings]
if not proposal.preview:
proposal.preview = f"{proposal.duration_s}s · {model.label}"
return ProducerAction(
payload=proposal,
preview=proposal.preview,
ask_user=proposal.ask_user,
)
def _is_vague_input(message: str) -> bool:
"""True, wenn die Eingabe keine entscheidenden Produktions-Infos enthält."""
normalized = message.casefold()
genre_terms = (
"riddim", "reggae", "dub", "techno", "club", "rave", "lofi", "lo-fi", "chill",
"film", "cinematic", "episch", "rock", "gitarre", "guitar", "pop", "hiphop",
"hip-hop", "house", "drum and bass", "dnb", "ambient", "jazz", "soul", "trap",
"edm", "trance", "metal", "punk", "blues", "funk", "disco", "country", "folk",
)
mood_terms = (
"dunkel", "dark", "euphorisch", "euphoric", "entspannt", "relaxed", "traurig",
"sad", "wütend", "angry", "aggressiv", "aggressive", "melancholisch", "melancholic",
"fröhlich", "happy", "warm", "kalt", "cold", "mysteriös", "mysterious", "episch",
"cinematic", "energetisch", "energetic", "ruhig", "calm", "böse", "evil", "düster",
)
instrument_terms = (
"bass", "drums", "drum", "gitarre", "guitar", "piano", "keys", "synth", "synthesizer",
"strings", "streicher", "violine", "saxophon", "sax", "horn", "flöte", "flute",
"percussion", "schlagzeug", "kick", "snare", "hi-hat", "pad", "lead", "vocal", "gesang",
"stimme", "choir", "chor", "organ", "orgel", "brass",
)
structure_terms = (
"instrumental", "vocal", "gesang", "lyrics", "text", "sänger", "sängerin",
"sekunden", "seconds", "minuten", "minutes", "bpm", "tempo", "dauer", "länge",
"intro", "verse", "chorus", "bridge", "drop", "hook",
)
has_info = any(
term in normalized
for term in (*genre_terms, *mood_terms, *instrument_terms, *structure_terms)
)
return not has_info
def _default_ask(message: str) -> str:
"""Deterministische Rückfrage, wenn die Eingabe zu wenig Info enthält."""
normalized = message.casefold()
if "instrumental" not in normalized and not any(
term in normalized for term in ("vocal", "gesang", "stimme", "lyrics", "text")
):
return "Soll es ein Instrumental sein oder mit Gesang?"
if not any(
term in normalized
for term in (
"bpm", "tempo", "sekunden", "seconds", "minuten", "minutes", "dauer", "länge",
)
):
return "Wie lang soll der Track sein (5–180 Sekunden)?"
return "Welche Stimmung soll der Track vermitteln (z. B. dunkel, euphorisch, entspannt)?"
def propose_fields(message: str) -> FieldProposal:
"""Compatibility helper using the validated local fallback/provider chain."""
return ProducerAgent(load_registry()).propose(message).payload
def propose_master(message: str) -> MasterAction:
if gemini_api_key():
try:
return _propose_master_llm(message)
except Exception:
pass # Fallback auf deterministische Heuristik
return _propose_master_fallback(message)
def _propose_master_llm(message: str) -> MasterAction:
"""Master-Vorschlag via Gemini; nutzt prompts/mastering.md als System-Prompt."""
api_key = gemini_api_key()
assert api_key is not None
model_id = os.environ.get("GEMINI_MODEL", "gemini-3.5-flash")
system_prompt = _read_prompt("MASTERING_SYSTEM_PROMPT", MASTER_PROMPT_PATH, _FALLBACK_MASTER)
base_url = "https://generativelanguage.googleapis.com/v1beta"
url = f"{base_url}/models/{model_id}:generateContent?key={api_key}"
body = {
"system_instruction": {"parts": [{"text": system_prompt}]},
"contents": [{"parts": [{"text": json.dumps({"idea": message}, ensure_ascii=False)}]}],
"generationConfig": {
"temperature": 0.2,
"maxOutputTokens": 2048,
"responseMimeType": "application/json",
},
}
request = urllib.request.Request(
url,
data=json.dumps(body).encode("utf-8"),
headers={"Content-Type": "application/json"},
method="POST",
)
try:
with urllib.request.urlopen(request, timeout=60) as response:
payload = json.loads(response.read().decode("utf-8"))
except urllib.error.HTTPError as error:
detail = error.read().decode("utf-8", errors="replace")[:300]
raise ValueError(f"Gemini HTTP {error.code}: {detail}") from error
except urllib.error.URLError as error:
raise ValueError(f"Gemini nicht erreichbar: {error.reason}") from error
candidates = payload.get("candidates") or []
if not candidates:
block = payload.get("promptFeedback", {}).get("blockReason")
raise ValueError(f"Gemini lieferte keine Antwort (blockReason={block}).")
text = candidates[0].get("content", {}).get("parts", [{}])[0].get("text", "")
cleaned = re.sub(r"^```(?:json)?\s*|\s*```$", "", text.strip())
parsed = json.loads(cleaned)
if not isinstance(parsed, dict):
raise ValueError("Gemini lieferte kein JSON-Objekt.")
payload_out = parsed.get("payload", parsed)
if not isinstance(payload_out, dict):
raise ValueError("Gemini lieferte kein gültiges Payload-Objekt.")
proposal = MasterProposal.model_validate(payload_out)
if proposal.purity_mode == "strict":
proposal.requires_strict_approval = True
return MasterAction(payload=proposal, preview=proposal.preview)
def _propose_master_fallback(message: str) -> MasterAction:
normalized = message.casefold()
analyze_only = any(
term in normalized for term in ("nur messen", "nur analyse", "only analyze", "analyse only")
)
if any(term in normalized for term in ("club", "dj", "dancefloor")):
profile = "club"
intensity = 72
elif any(term in normalized for term in ("lauter", "loud", "druck")):
profile = "streaming_loud"
intensity = 65
elif any(term in normalized for term in ("hifi", "dynamisch", "audiophil")):
profile = "dynamic_hifi"
intensity = 42
else:
profile = "streaming_safe"
intensity = 50
strict = "strict" in normalized
purity_off = any(
term in normalized for term in ("ohne purity", "kein purity", "nur master", "skip purity")
)
purity_mode = "strict" if strict else "off" if purity_off else "soft_keep_air"
preserve_air = any(
term in normalized for term in ("luft", "air", "wärme", "hall", "nicht tot")
)
keep_air_db = -24.0 if preserve_air else -20.0
margin = 1.5 if preserve_air else 2.0
rationale = (
"Strict wurde nur als Vorschlag markiert und benötigt eine separate Bestätigung."
if strict
else "soft_keep_air schützt Air, Hall und Wärme; das Original bleibt unverändert."
)
if analyze_only:
rationale = "Es wird ausschließlich gemessen; keine Audiodatei wird verändert oder erzeugt."
proposal = MasterProposal(
profile=profile,
purity_mode=purity_mode,
keep_air_db=keep_air_db,
margin=margin,
intensity=intensity,
analyze_only=analyze_only,
rationale=rationale,
preview=f"{profile} · {purity_mode}{' · nur Analyse' if analyze_only else ''}",
requires_strict_approval=strict,
)
return MasterAction(payload=proposal, preview=proposal.preview)
|