File size: 12,680 Bytes
842a99f 8ba62a7 842a99f 2446a8b 842a99f dd42331 2446a8b dd42331 2446a8b dd42331 2446a8b dd42331 2446a8b dd42331 2446a8b dd42331 2446a8b dd42331 842a99f dd42331 842a99f dd42331 842a99f dd42331 7457794 dd42331 842a99f 78cac08 6d5b723 78cac08 6d5b723 7ae8a62 78cac08 6d5b723 78cac08 6d5b723 78cac08 7ae8a62 6d5b723 78cac08 6d5b723 78cac08 e87e83d 7ae8a62 78cac08 e87e83d 7ae8a62 78cac08 7ae8a62 6d5b723 7ae8a62 e87e83d 7ae8a62 e87e83d 7ae8a62 78cac08 6d5b723 78cac08 6d5b723 1355fb6 78cac08 |
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 |
"""
One-shot music generation functions for MagentaRT.
This module contains the core generation functions extracted from the main app
that can be used independently for single-shot music generation tasks.
"""
import math
import numpy as np
from magenta_rt import audio as au
from utils import (
match_loudness_to_reference,
stitch_generated,
hard_trim_seconds,
apply_micro_fades,
make_bar_aligned_context,
take_bar_aligned_tail
)
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,
loudness_mode: str = "auto",
loudness_headroom_db: float = 1.0,
intro_bars_to_drop: int = 0,
progress_cb=None
):
"""
Generate a continuation of an input loop using MagentaRT.
"""
# ===== CRITICAL FIX: Force codec state isolation =====
# Create a completely isolated encoding session to prevent
# audio from previous generations bleeding into this one
# Save original codec state (if any)
original_codec_state = {}
codec_attrs_to_clear = [
'_encode_state', '_decode_state',
'_last_encoded', '_last_decoded',
'_encoder_cache', '_decoder_cache',
'_buffer', '_frame_buffer'
]
for attr in codec_attrs_to_clear:
if hasattr(mrt.codec, attr):
original_codec_state[attr] = getattr(mrt.codec, attr)
setattr(mrt.codec, attr, None)
# Also clear any MRT-level generation state
mrt_attrs_to_clear = ['_last_state', '_generation_cache']
for attr in mrt_attrs_to_clear:
if hasattr(mrt, attr):
original_codec_state[f'mrt_{attr}'] = getattr(mrt, attr)
setattr(mrt, attr, None)
try:
# ============================================================
# Load & prep - Force FRESH file read (no caching)
loop = au.Waveform.from_file(input_wav_path).resample(mrt.sample_rate).as_stereo()
# CRITICAL: Create a detached copy to prevent reference issues
loop = au.Waveform(
loop.samples.copy(), # Force array copy
loop.sample_rate
)
# Use tail for context
codec_fps = float(mrt.codec.frame_rate)
ctx_seconds = float(mrt.config.context_length_frames) / codec_fps
loop_for_context = take_bar_aligned_tail(loop, bpm, beats_per_bar, ctx_seconds)
# CRITICAL: Another detached copy before encoding
loop_for_context = au.Waveform(
loop_for_context.samples.copy(),
loop_for_context.sample_rate
)
# Force fresh encoding with explicit copy flags
tokens_full = mrt.codec.encode(loop_for_context).astype(np.int32, copy=True)
tokens = tokens_full[:, :mrt.config.decoder_codec_rvq_depth]
# CRITICAL: Ensure tokens are not a view
tokens = np.array(tokens, dtype=np.int32, copy=True, order='C')
# Bar-aligned token window
context_tokens = make_bar_aligned_context(
tokens, bpm=bpm, fps=float(mrt.codec.frame_rate),
ctx_frames=mrt.config.context_length_frames, beats_per_bar=beats_per_bar
)
# CRITICAL: Force contiguous memory layout
context_tokens = np.ascontiguousarray(context_tokens, dtype=np.int32)
# Create completely fresh state
state = mrt.init_state()
state.context_tokens = context_tokens
# STYLE embed - force fresh
loop_embed = mrt.embed_style(loop_for_context)
embeds, weights = [np.array(loop_embed, copy=True)], [float(loop_weight)]
if extra_styles:
for i, s in enumerate(extra_styles):
if s.strip():
e = mrt.embed_style(s.strip())
embeds.append(np.array(e, copy=True))
w = style_weights[i] if (style_weights and i < len(style_weights)) else 1.0
weights.append(float(w))
wsum = float(sum(weights)) or 1.0
weights = [w / wsum for w in weights]
combined_style = np.sum([w * e for w, e in zip(weights, embeds)], axis=0)
combined_style = np.ascontiguousarray(combined_style, dtype=np.float32)
# --- Length math ---
seconds_per_bar = beats_per_bar * (60.0 / bpm)
total_secs = bars * seconds_per_bar
drop_bars = max(0, int(intro_bars_to_drop))
drop_secs = min(drop_bars, bars) * seconds_per_bar
gen_total_secs = total_secs + drop_secs
chunk_secs = mrt.config.chunk_length_frames * mrt.config.frame_length_samples / mrt.sample_rate
steps = int(math.ceil(gen_total_secs / chunk_secs)) + 1
if progress_cb:
progress_cb(0, steps)
# Generate with state isolation
chunks = []
for i in range(steps):
wav, state = mrt.generate_chunk(state=state, style=combined_style)
# Force copy the waveform samples to prevent reference issues
wav = au.Waveform(wav.samples.copy(), wav.sample_rate)
chunks.append(wav)
if progress_cb:
progress_cb(i + 1, steps)
# Rest unchanged...
stitched = stitch_generated(chunks, mrt.sample_rate, mrt.config.crossfade_length).as_stereo()
stitched = hard_trim_seconds(stitched, gen_total_secs)
if drop_secs > 0:
n_drop = int(round(drop_secs * stitched.sample_rate))
stitched = au.Waveform(stitched.samples[n_drop:], stitched.sample_rate)
out = hard_trim_seconds(stitched, total_secs)
out, loud_stats = apply_barwise_loudness_match(
out=out,
ref_loop=loop,
bpm=bpm,
beats_per_bar=beats_per_bar,
method=loudness_mode,
headroom_db=loudness_headroom_db,
smooth_ms=50,
)
apply_micro_fades(out, 5)
return out, loud_stats
finally:
# ===== CLEANUP: Clear codec state after generation =====
# This prevents audio from THIS generation leaking into the NEXT one
for attr in codec_attrs_to_clear:
if hasattr(mrt.codec, attr):
setattr(mrt.codec, attr, None)
for attr in mrt_attrs_to_clear:
if hasattr(mrt, attr):
setattr(mrt, attr, None)
# =======================================================
def generate_style_only_with_mrt(
mrt,
bpm: float,
bars: int = 8,
beats_per_bar: int = 4,
styles: str = "warmup",
style_weights: str = "",
intro_bars_to_drop: int = 0,
):
"""
Style-only, bar-aligned generation using a silent context (no input audio).
Returns: (au.Waveform out, dict loud_stats_or_None)
"""
# ---- Build a 10s silent context, tokenized for the model ----
codec_fps = float(mrt.codec.frame_rate)
ctx_seconds = float(mrt.config.context_length_frames) / codec_fps
sr = int(mrt.sample_rate)
silent = au.Waveform(np.zeros((int(round(ctx_seconds * sr)), 2), np.float32), sr)
tokens_full = mrt.codec.encode(silent).astype(np.int32)
tokens = tokens_full[:, :mrt.config.decoder_codec_rvq_depth]
state = mrt.init_state()
state.context_tokens = tokens
# ---- Style vector (text prompts only, normalized weights) ----
prompts = [s.strip() for s in (styles.split(",") if styles else []) if s.strip()]
if not prompts:
prompts = ["warmup"]
sw = [float(x) for x in style_weights.split(",")] if style_weights else []
embeds, weights = [], []
for i, p in enumerate(prompts):
embeds.append(mrt.embed_style(p))
weights.append(sw[i] if i < len(sw) else 1.0)
wsum = float(sum(weights)) or 1.0
weights = [w / wsum for w in weights]
style_vec = np.sum([w * e for w, e in zip(weights, embeds)], axis=0).astype(np.float32)
# ---- Target length math ----
seconds_per_bar = beats_per_bar * (60.0 / bpm)
total_secs = bars * seconds_per_bar
drop_bars = max(0, int(intro_bars_to_drop))
drop_secs = min(drop_bars, bars) * seconds_per_bar
gen_total_secs = total_secs + drop_secs
# ~2.0s chunk length from model config
chunk_secs = (mrt.config.chunk_length_frames * mrt.config.frame_length_samples) / float(mrt.sample_rate)
# Generate enough chunks to cover total, plus a pad chunk for crossfade headroom
steps = int(math.ceil(gen_total_secs / chunk_secs)) + 1
chunks = []
for _ in range(steps):
wav, state = mrt.generate_chunk(state=state, style=style_vec)
chunks.append(wav)
# Stitch & trim to exact musical length
stitched = stitch_generated(chunks, mrt.sample_rate, mrt.config.crossfade_length).as_stereo()
stitched = hard_trim_seconds(stitched, gen_total_secs)
if drop_secs > 0:
n_drop = int(round(drop_secs * stitched.sample_rate))
stitched = au.Waveform(stitched.samples[n_drop:], stitched.sample_rate)
out = hard_trim_seconds(stitched, total_secs)
out = out.peak_normalize(0.95)
apply_micro_fades(out, 5)
return out, None # loudness stats not applicable (no reference)
# loudness matching helper for /generate:
def apply_barwise_loudness_match(
out: au.Waveform,
ref_loop: au.Waveform,
*,
bpm: float,
beats_per_bar: int,
method: str = "auto",
headroom_db: float = 1.0,
smooth_ms: int = 50,
) -> tuple[au.Waveform, dict]:
"""
Bar-locked loudness matching that establishes the correct starting level
then maintains consistency. Only the first bar is matched to the reference;
subsequent bars use the same gain to maintain relative dynamics.
"""
sr = int(out.sample_rate)
spb = (60.0 / float(bpm)) * int(beats_per_bar)
bar_len = int(round(spb * sr))
y = out.samples.astype(np.float32, copy=False)
if y.ndim == 1: y = y[:, None]
if ref_loop.sample_rate != sr:
ref = ref_loop.resample(sr).as_stereo().samples.astype(np.float32, copy=False)
else:
ref = ref_loop.as_stereo().samples.astype(np.float32, copy=False)
if ref.ndim == 1: ref = ref[:, None]
if ref.shape[1] == 1: ref = np.repeat(ref, 2, axis=1)
from utils import match_loudness_to_reference
# Measure reference loudness once
ref_bar_len = min(ref.shape[0], bar_len)
ref_bar = au.Waveform(ref[:ref_bar_len], sr)
gains_db = []
out_adj = y.copy()
need = y.shape[0]
n_bars = max(1, int(np.ceil(need / float(bar_len))))
ramp = int(max(0, round(smooth_ms * sr / 1000.0)))
min_lufs_samples = int(0.4 * sr)
# Calculate gain from bar 0 matching
first_bar_gain_linear = 1.0
for i in range(n_bars):
s = i * bar_len
e = min(need, s + bar_len)
if e <= s:
break
bar_samples = e - s
tgt_bar = au.Waveform(y[s:e], sr) # Always read from ORIGINAL
# First bar: match to reference to establish gain
if i == 0:
effective_method = "rms" if bar_samples < min_lufs_samples else method
matched_bar, stats = match_loudness_to_reference(
ref_bar, tgt_bar, method=effective_method, headroom_db=headroom_db
)
# Calculate the linear gain that was applied
eps = 1e-12
first_bar_gain_linear = float(np.sqrt(
(np.mean(matched_bar.samples**2) + eps) /
(np.mean(tgt_bar.samples**2) + eps)
))
g = matched_bar.samples.astype(np.float32, copy=False)
else:
# Subsequent bars: apply the same gain from bar 0
g = (tgt_bar.samples * first_bar_gain_linear).astype(np.float32, copy=False)
# Calculate gain in dB for stats
if tgt_bar.samples.size > 0:
eps = 1e-12
g_lin = float(np.sqrt((np.mean(g**2) + eps) / (np.mean(tgt_bar.samples**2) + eps)))
else:
g_lin = 1.0
gains_db.append(20.0 * np.log10(max(g_lin, 1e-6)))
# Apply with ramp for smoothness
if i > 0 and ramp > 0:
ramp_len = min(ramp, e - s)
t = np.linspace(0.0, 1.0, ramp_len, dtype=np.float32)[:, None]
out_adj[s:s+ramp_len] = (1.0 - t) * out_adj[s:s+ramp_len] + t * g[:ramp_len]
out_adj[s+ramp_len:e] = g[ramp_len:e-s]
else:
out_adj[s:e] = g
out.samples = out_adj.astype(np.float32, copy=False)
return out, {"per_bar_gain_db": gains_db} |