Spaces:
Running
Running
File size: 13,787 Bytes
f70477a ede0049 f8b3793 ede0049 f70477a 241e975 f70477a 241e975 f70477a 241e975 f70477a ede0049 f8b3793 f70477a ede0049 f70477a f8b3793 8cf69d0 f8b3793 8cf69d0 f8b3793 8cf69d0 f8b3793 f70477a f8b3793 f70477a f8b3793 f70477a f8b3793 f70477a |
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 |
from magenta_rt import system, audio as au
import numpy as np
from fastapi import FastAPI, UploadFile, File, Form
import tempfile, io, base64, math, threading
from fastapi.middleware.cors import CORSMiddleware
from contextlib import contextmanager
import soundfile as sf
import numpy as np
from math import gcd
from scipy.signal import resample_poly
@contextmanager
def mrt_overrides(mrt, **kwargs):
"""Temporarily set attributes on MRT if they exist; restore after."""
old = {}
try:
for k, v in kwargs.items():
if hasattr(mrt, k):
old[k] = getattr(mrt, k)
setattr(mrt, k, v)
yield
finally:
for k, v in old.items():
setattr(mrt, k, v)
# loudness utils
try:
import pyloudnorm as pyln
_HAS_LOUDNORM = True
except Exception:
_HAS_LOUDNORM = False
def _measure_lufs(wav: au.Waveform) -> float:
# pyloudnorm expects float32/float64, shape (n,) or (n, ch)
meter = pyln.Meter(wav.sample_rate) # defaults to BS.1770-4
return float(meter.integrated_loudness(wav.samples))
def _rms(x: np.ndarray) -> float:
if x.size == 0: return 0.0
return float(np.sqrt(np.mean(x**2)))
def match_loudness_to_reference(
ref: au.Waveform,
target: au.Waveform,
method: str = "auto", # "auto"|"lufs"|"rms"|"none"
headroom_db: float = 1.0
) -> tuple[au.Waveform, dict]:
"""
Scales `target` to match `ref` loudness. Returns (adjusted_wave, stats).
"""
stats = {"method": method, "applied_gain_db": 0.0}
if method == "none":
return target, stats
if method == "auto":
method = "lufs" if _HAS_LOUDNORM else "rms"
if method == "lufs" and _HAS_LOUDNORM:
L_ref = _measure_lufs(ref)
L_tgt = _measure_lufs(target)
delta_db = L_ref - L_tgt
gain = 10.0 ** (delta_db / 20.0)
y = target.samples.astype(np.float32) * gain
stats.update({"ref_lufs": L_ref, "tgt_lufs_before": L_tgt, "applied_gain_db": delta_db})
else:
# RMS fallback
ra = _rms(ref.samples)
rb = _rms(target.samples)
if rb <= 1e-12:
return target, stats
gain = ra / rb
y = target.samples.astype(np.float32) * gain
stats.update({"ref_rms": ra, "tgt_rms_before": rb, "applied_gain_db": 20*np.log10(max(gain,1e-12))})
# simple peak “limiter” to keep headroom
limit = 10 ** (-headroom_db / 20.0) # e.g., -1 dBFS
peak = float(np.max(np.abs(y))) if y.size else 0.0
if peak > limit:
y *= (limit / peak)
stats["post_peak_limited"] = True
else:
stats["post_peak_limited"] = False
target.samples = y.astype(np.float32)
return target, stats
# ----------------------------
# Crossfade stitch (your good path)
# ----------------------------
def stitch_generated(chunks, sr, xfade_s):
if not chunks:
raise ValueError("no chunks")
xfade_n = int(round(xfade_s * sr))
if xfade_n <= 0:
return au.Waveform(np.concatenate([c.samples for c in chunks], axis=0), sr)
t = np.linspace(0, np.pi/2, xfade_n, endpoint=False, dtype=np.float32)
eq_in, eq_out = np.sin(t)[:, None], np.cos(t)[:, None]
first = chunks[0].samples
if first.shape[0] < xfade_n:
raise ValueError("chunk shorter than crossfade prefix")
out = first[xfade_n:].copy() # drop model pre-roll
for i in range(1, len(chunks)):
cur = chunks[i].samples
if cur.shape[0] < xfade_n:
continue
head, tail = cur[:xfade_n], cur[xfade_n:]
mixed = out[-xfade_n:] * eq_out + head * eq_in
out = np.concatenate([out[:-xfade_n], mixed, tail], axis=0)
return au.Waveform(out, sr)
# ----------------------------
# Bar-aligned token context
# ----------------------------
def make_bar_aligned_context(tokens, bpm, fps=25, ctx_frames=250, beats_per_bar=4):
frames_per_bar_f = (beats_per_bar * 60.0 / bpm) * fps
frames_per_bar = int(round(frames_per_bar_f))
if abs(frames_per_bar - frames_per_bar_f) > 1e-3:
reps = int(np.ceil(ctx_frames / len(tokens)))
return np.tile(tokens, (reps, 1))[-ctx_frames:]
reps = int(np.ceil(ctx_frames / len(tokens)))
tiled = np.tile(tokens, (reps, 1))
end = (len(tiled) // frames_per_bar) * frames_per_bar
if end < ctx_frames:
return tiled[-ctx_frames:]
start = end - ctx_frames
return tiled[start:end]
def hard_trim_seconds(wav: au.Waveform, seconds: float) -> au.Waveform:
n = int(round(seconds * wav.sample_rate))
return au.Waveform(wav.samples[:n], wav.sample_rate)
def apply_micro_fades(wav: au.Waveform, ms: int = 5) -> None:
n = int(wav.sample_rate * ms / 1000.0)
if n > 0 and wav.samples.shape[0] > 2*n:
env = np.linspace(0.0, 1.0, n, dtype=np.float32)[:, None]
wav.samples[:n] *= env
wav.samples[-n:] *= env[::-1]
def take_bar_aligned_tail(wav: au.Waveform,
bpm: float,
beats_per_bar: int,
ctx_seconds: float) -> au.Waveform:
"""
Return the LAST N bars whose duration is as close as possible to ctx_seconds,
anchored to the end of `wav`, and bar-aligned.
"""
spb = (60.0 / bpm) * beats_per_bar # seconds per bar
bars_needed = max(1, int(round(ctx_seconds / spb)))
tail_seconds = bars_needed * spb # exact multiple of bars
n = int(round(tail_seconds * wav.sample_rate))
if n >= wav.samples.shape[0]:
# Input shorter than desired tail: keep whole thing (your existing behavior will tile)
return wav
return au.Waveform(wav.samples[-n:], wav.sample_rate)
# ----------------------------
# Main generation (single combined style vector)
# ----------------------------
def generate_loop_continuation_with_mrt(
mrt,
input_wav_path: str,
bpm: float,
extra_styles=None,
style_weights=None,
bars: int = 8,
beats_per_bar: int = 4,
loop_weight: float = 1.0, # NEW
loudness_mode: str = "auto", # "auto"|"lufs"|"rms"|"none"
loudness_headroom_db: float = 1.0, # for the peak guard
):
# Load loop & put into model SR/channels
loop = au.Waveform.from_file(input_wav_path).resample(mrt.sample_rate).as_stereo()
# Compute the model's desired context seconds (e.g., 250 frames / 25 fps = 10s)
codec_fps = float(mrt.codec.frame_rate)
ctx_seconds = float(mrt.config.context_length_frames) / codec_fps # typically 10.0s
# ✅ NEW: take bar-aligned TAIL for context, if input is long enough
loop_for_context = take_bar_aligned_tail(
wav=loop,
bpm=bpm,
beats_per_bar=beats_per_bar,
ctx_seconds=ctx_seconds
)
# Encode ONLY the tail (so we condition on recent audio)
tokens_full = mrt.codec.encode(loop_for_context).astype(np.int32)
tokens = tokens_full[:, :mrt.config.decoder_codec_rvq_depth]
# Context
context_tokens = make_bar_aligned_context(
tokens,
bpm=bpm,
fps=int(mrt.codec.frame_rate),
ctx_frames=mrt.config.context_length_frames,
beats_per_bar=beats_per_bar,
)
state = mrt.init_state()
state.context_tokens = context_tokens
# ---------- STYLE: weighted avg into ONE vector ----------
# Base embed from loop with adjustable loop_weight
embeds = []
weights = []
# loop embedding
loop_embed = mrt.embed_style(loop)
embeds.append(loop_embed)
weights.append(float(loop_weight)) # <--- use requested loop weight
# extra styles
if extra_styles:
for i, s in enumerate(extra_styles):
if s.strip():
embeds.append(mrt.embed_style(s.strip()))
w = style_weights[i] if (style_weights and i < len(style_weights)) else 1.0
weights.append(float(w))
# Prevent all-zero weights; normalize
wsum = float(sum(weights))
if wsum <= 0.0:
# fallback: rely on loop to avoid NaNs
weights = [1.0] + [0.0] * (len(weights) - 1)
wsum = 1.0
weights = [w / wsum for w in weights]
# weighted sum -> single style vector (match dtype)
combined_style = np.sum([w * e for w, e in zip(weights, embeds)], axis=0).astype(loop_embed.dtype)
# Chunks to cover exact bars
seconds_per_bar = beats_per_bar * (60.0 / bpm)
total_secs = bars * seconds_per_bar
chunk_secs = mrt.config.chunk_length_frames * mrt.config.frame_length_samples / mrt.sample_rate # ~2.0
steps = int(math.ceil(total_secs / chunk_secs)) + 1 # pad then trim
# Generate
chunks = []
for _ in range(steps):
wav, state = mrt.generate_chunk(state=state, style=combined_style) # ONE style vector
chunks.append(wav)
# Stitch -> trim -> polish
out = stitch_generated(chunks, mrt.sample_rate, mrt.config.crossfade_length).as_stereo()
out = hard_trim_seconds(out, total_secs).peak_normalize(0.95)
apply_micro_fades(out, 5)
# Loudness match to the *input loop* so the return level feels consistent
out, loud_stats = match_loudness_to_reference(
ref=loop, target=out,
method=loudness_mode,
headroom_db=loudness_headroom_db,
)
return out, loud_stats
# ----------------------------
# FastAPI app with lazy, thread-safe model init
# ----------------------------
app = FastAPI()
app.add_middleware(
CORSMiddleware,
allow_origins=["*"], # or lock to your domain(s)
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
_MRT = None
_MRT_LOCK = threading.Lock()
def get_mrt():
global _MRT
if _MRT is None:
with _MRT_LOCK:
if _MRT is None:
_MRT = system.MagentaRT(tag="base", guidance_weight=1.0, device="gpu", lazy=False)
return _MRT
@app.post("/generate")
def generate(
loop_audio: UploadFile = File(...),
bpm: float = Form(...),
bars: int = Form(8),
beats_per_bar: int = Form(4),
styles: str = Form("acid house"),
style_weights: str = Form(""),
loop_weight: float = Form(1.0),
loudness_mode: str = Form("auto"),
loudness_headroom_db: float = Form(1.0),
guidance_weight: float = Form(5.0),
temperature: float = Form(1.1),
topk: int = Form(40),
target_sample_rate: int | None = Form(None), # <-- add this
):
# Read file
data = loop_audio.file.read()
if not data:
return {"error": "Empty file"}
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
tmp.write(data)
tmp_path = tmp.name
# Parse styles + weights
extra_styles = [s for s in (styles.split(",") if styles else []) if s.strip()]
weights = [float(x) for x in style_weights.split(",")] if style_weights else None
mrt = get_mrt() # warm once, in this worker thread
# Temporarily override MRT inference knobs for this request
with mrt_overrides(mrt,
guidance_weight=guidance_weight,
temperature=temperature,
topk=topk):
wav, loud_stats = generate_loop_continuation_with_mrt(
mrt,
input_wav_path=tmp_path,
bpm=bpm,
extra_styles=extra_styles,
style_weights=weights,
bars=bars,
beats_per_bar=beats_per_bar,
loop_weight=loop_weight,
loudness_mode=loudness_mode,
loudness_headroom_db=loudness_headroom_db,
)
# 1) Figure out the desired SR
inp_info = sf.info(tmp_path)
input_sr = int(inp_info.samplerate)
target_sr = int(target_sample_rate or input_sr)
# 2) Convert magenta output to target_sr if needed
# wav.samples: shape [num_samples, num_channels], float32/-1..1 (per your code)
cur_sr = int(mrt.sample_rate)
x = wav.samples # np.ndarray (S, C)
if cur_sr != target_sr:
g = gcd(cur_sr, target_sr)
up, down = target_sr // g, cur_sr // g
# ensure 2D shape (S, C)
x = wav.samples
if x.ndim == 1:
x = x[:, None]
y = np.column_stack([resample_poly(x[:, ch], up, down) for ch in range(x.shape[1])])
else:
y = wav.samples if wav.samples.ndim == 2 else wav.samples[:, None]
# 3) Snap to exact frame count for loop-perfect length
seconds_per_bar = (60.0 / float(bpm)) * int(beats_per_bar)
expected_len = int(round(float(bars) * seconds_per_bar * target_sr))
if y.shape[0] < expected_len:
pad = np.zeros((expected_len - y.shape[0], y.shape[1]), dtype=y.dtype)
y = np.vstack([y, pad])
elif y.shape[0] > expected_len:
y = y[:expected_len, :]
total_samples = int(y.shape[0])
loop_duration_seconds = total_samples / float(target_sr)
# 4) Write y into buf as WAV @ target_sr
buf = io.BytesIO()
sf.write(buf, y, target_sr, subtype="FLOAT", format="WAV")
buf.seek(0)
audio_b64 = base64.b64encode(buf.read()).decode("utf-8")
# 5) Update metadata to be authoritative
metadata = {
"bpm": int(round(bpm)),
"bars": int(bars),
"beats_per_bar": int(beats_per_bar),
"styles": extra_styles,
"style_weights": weights,
"loop_weight": loop_weight,
"loudness": loud_stats,
"sample_rate": int(target_sr),
"channels": int(y.shape[1]),
"crossfade_seconds": mrt.config.crossfade_length,
"total_samples": total_samples,
"seconds_per_bar": seconds_per_bar,
"loop_duration_seconds": loop_duration_seconds,
"guidance_weight": guidance_weight,
"temperature": temperature,
"topk": topk,
}
return {"audio_base64": audio_b64, "metadata": metadata}
@app.get("/health")
def health():
return {"ok": True} |