SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.09901 · ICLR · 2024

Whole-Song Hierarchical Generation of Symbolic Music Using Cascaded Diffusion Models

Ziyu Wang, Gus Xia

arXiv · PDF · Open in the Atlas

Code that ran

10 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 9 failures of our instrument; 2 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2405.09901.json · badge: /api/ran/2405.09901.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
zzwaang/melody-reduction-algo canonical (classifier label)
classifier basis: census
9 ran · 2 no recorded run
zzwaang/whole-song-gen canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
args_setting_to_fn Ran
ran; our contract draft was wrong, not the code
zzwaang/whole-song-gen/train_main.py
code served (permissive licence) · get_code("7fcbca19e574093c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

class Args:
    def __init__(self):
        self.mode = 'train'
        self.autoreg = True
        self.external = False
        self.mask_bg = True
        self.multi_label = False
        self.uniform_pitch_shift = True
        self.debug = False

ARGS = {'args': Args()}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
chord_tone_analysis Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/preprocess.py
code served (permissive licence) · get_code("ec1fe56f43607473")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 2, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct small but valid arguments for chord_tone_analysis

# note_mat: shape (n_note, 2+) where column 1 contains pitch values
# We'll create 4 notes with pitches in column 1
note_mat = np.array([
    [0, 0, 100],   # note 0, pitch 0
    [1, 5, 101],   # note 1, pitch 5
    [2, 7, 102],   # note 2, pitch 7
    [3, 11, 103]   # note 3, pitch 11
], dtype=np.float64)

# chord_mat: shape (n_chord, 2+14) where columns 2-13 represent pitch classes 0-11
# Columns 0-1 are metadata, columns 2-13 are one-hot encodings for pitches 0-11
# We'll create 3 chords
# Chord 0: contains pitches 0, 4, 7 (C major)
# Chord 1: contains pitches 0, 3, 7 (C minor)
# Chord 2: contains pitches 5, 9, 0 (F major)
chord_mat = np.zeros((3, 14), dtype=np.int64)
chord_mat[0, [2, 6, 9]] = 1      # chord 0: pitches 0, 4, 7
chord_mat[1, [2, 5, 9]] = 1      # chord 1: pitches 0, 3, 7
chord_mat[2, [7, 11, 2]] = 1     # chord 2: pitches 5, 9, 0

# chord_ids: array mapping each note to a chord index
# 4 notes mapped to 3 chords
chord_ids = np.array([0, 0, 1, 2], dtype=np.int64)

ARGS = {
    "note_mat": note_mat,
    "chord_mat": chord_mat,
    "chord_ids": chord_ids
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_chord_density Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/postprocess.py
code served (permissive licence) · get_code("28ca3ef943791351")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Analyze the function to understand what's needed:
# - path: an iterable of indices (used as: for pid, p in enumerate(path), note_mat[p])
# - note_mat: indexable object where note_mat[p] returns a note with at least indices [5] and [6]
# - chords: list of chords where each chord has a [-1] element (max_density)

# Create a small path (list of indices)
path = [0, 1, 0, 2, 1]

# Create note_mat as a list of notes
# Each note needs at least indices 5 and 6
# note[6] should be a chord_id, note[5] is a fallback chord_id
# Let's make notes with 7 elements (indices 0-6)
note_mat = [
    [0, 0, 0, 0, 0, 0, 0],  # note 0: chord_id=0 (at index 6), fallback=0 (at index 5)
    [0, 0, 0, 0, 0, 1, 1],  # note 1: chord_id=1 (at index 6), fallback=1 (at index 5)
    [0, 0, 0, 0, 0, 2, 0],  # note 2: chord_id=0 (at index 6), fallback=2 (at index 5)
]

# Create chords as a list where each chord has a max_density at index [-1]
# We need at least 3 chords (for chord_ids 0, 1, 2)
chords = [
    [1, 2, 3, 5],    # chord 0: max_density = 5
    [1, 2, 3, 4],    # chord 1: max_density = 4
    [1, 2, 3, 3],    # chord 2: max_density = 3
]

ARGS = {
    "path": path,
    "note_mat": note_mat,
    "chords": chords,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_onset_type Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/shortest_path_algo.py
code served (permissive licence) · get_code("c6a39c23b51cf3a7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family int, shape (7,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a small onset array with various values that will test different conditions
# The function classifies onset times based on their divisibility by different note lengths
onset = np.array([0, 1, 2, 4, 8, 12, 16], dtype=np.int64)

ARGS = {
    "onset": onset,
    "nbpm": 4,
    "nspb": 4,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (7,) — shape (7,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
path_to_chord_bins Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/postprocess.py
code served (permissive licence) · get_code("f4759550e29f75c4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Analyze the function to understand what's needed:
# - path: dict with 'path' (list of indices) and 'reduction_rate' (float)
# - note_mat: 2D array-like, indexed with [p] and [p, 2], needs columns 2, 5, 6
# - chord: 2D array-like, indexed with [i, int(note[2] % 12) + 1]

# note_mat structure:
# - Indexed by single index [p]: returns a note (needs indices 2, 5, 6)
# - Indexed by double index [p, 2]: returns a value
# - note[2] is used for modulo 12, so it's a pitch-like value
# - note[5] and note[6] are chord IDs
# We'll create a 2D array where each row is a note with at least 7 columns

# Create note_mat: shape (10, 7) - 10 notes, 7 attributes
note_mat = np.array([
    [0, 0, 0, 0, 0, 0, 0],   # note 0
    [0, 0, 2, 0, 0, 1, 1],   # note 1
    [0, 0, 4, 0, 0, 1, 1],   # note 2
    [0, 0, 5, 0, 0, 0, 0],   # note 3
    [0, 0, 7, 0, 0, 1, 1],   # note 4
    [0, 0, 9, 0, 0, 2, 2],   # note 5
    [0, 0, 11, 0, 0, 2, 2],  # note 6
    [0, 0, 0, 0, 0, 1, 1],   # note 7
    [0, 0, 2, 0, 0, 0, 0],   # note 8
    [0, 0, 4, 0, 0, 2, 2],   # note 9
], dtype=np.int32)

# Create path dict: indices into note_mat
path_dict = {
    'path': [0, 1, 2, 3, 4, 5, 6, 7, 8, 9],  # list of indices
    'reduction_rate': 0.8
}

# Create chord: shape (3, 13) - 3 chords, 13 columns (0-12 for pitch classes)
# chord[i, pitch_class + 1] indicates if that pitch is in chord i
chord = np.array([
    [0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0],  # chord 0: C major (C, E, G)
    [0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 0],  # chord 1: C major + extra
    [0, 1, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0],  # chord 2: C, E, A
], dtype=np.int32)

ARGS = {
    'path': path_dict,
    'note_mat': note_mat,
    'chord': chord
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_label Ran
ran; our contract draft was wrong, not the code
zzwaang/melody-reduction-algo/utils/read_pop909.py
code served (permissive licence) · get_code("433e78e1dce73946")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os

# Create a temporary file with a valid phrase label
temp_dir = tempfile.mkdtemp()
label_file = os.path.join(temp_dir, "human_label1.txt")

# Write a valid phrase label string
# Format: letter followed by duration (e.g., "A32B16C8" means A phrase of 32 frames, B of 16, C of 8)
phrase_label_content = "A32B16C8D24"

with open(label_file, 'w') as f:
    f.write(phrase_label_content)

ARGS = {
    'label_fn': label_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
read_melody Ran
ran, honoured the contract we drafted
zzwaang/melody-reduction-algo/utils/read_pop909.py
code served (permissive licence) · get_code("598a1dd94153ebfc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family int, shape (4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import tempfile
import os

# Create a temporary melody.txt file with valid content
temp_dir = tempfile.mkdtemp()
melody_fn = os.path.join(temp_dir, "melody.txt")

# Write a valid melody file
# Format: each line is "pitch duration"
# pitch can be 0 (rest) or non-zero (note)
# duration is a positive integer
melody_content = """60 4
0 2
62 3
64 2
0 1
65 3"""

with open(melody_fn, 'w') as f:
    f.write(melody_content)

ARGS = {
    "melody_fn": melody_fn
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family int, shape (4, 3) — shape (4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
remove_offset Ran
ran; our contract draft was wrong, not the code
zzwaang/melody-reduction-algo/tonal_reduction_algo/preprocess.py
code served (permissive licence) · get_code("1a4ffbef4040544c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create sample note matrix: shape (5, 2) where columns are [step, pitch]
note_mat = np.array([
    [10, 60],
    [15, 64],
    [20, 67],
    [25, 72],
    [30, 76]
], dtype=np.float32)

# Create sample chord matrix: shape (4, 2) where columns are [beat, chord_id]
chord_mat = np.array([
    [2, 0],
    [4, 1],
    [6, 2],
    [8, 3]
], dtype=np.float32)

# Define start_measure
start_measure = 2

# Define measure_to_step_fn: converts measure number to step number
def measure_to_step_fn(measure):
    return measure * 4  # 4 steps per measure

# Define measure_to_beat_fn: converts measure number to beat number
def measure_to_beat_fn(measure):
    return measure * 2  # 2 beats per measure

ARGS = {
    "note_mat": note_mat,
    "chord_mat": chord_mat,
    "start_measure": start_measure,
    "measure_to_step_fn": measure_to_step_fn,
    "measure_to_beat_fn": measure_to_beat_fn
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
return_chord_tone_score Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/shortest_path_algo.py
code served (permissive licence) · get_code("c3758ae46c228a29")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct a note_matrix with the required shape
# The function accesses note_matrix[:, 7], so we need at least 8 columns (0-7)
# Create a matrix with shape (2, 8) where column 7 contains binary values (0 or 1)
note_matrix = np.array([
    [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 1],
    [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0]
], dtype=np.float32)

ARGS = {
    "note_matrix": note_matrix
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float — no output shape is stated: observe_signature.py could not observe it (import:TypeError)
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
return_onset_score Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zzwaang/melody-reduction-algo/tonal_reduction_algo/shortest_path_algo.py
code served (permissive licence) · get_code("17e3972e3dd8279b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (5,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a note_matrix with shape (num_notes, num_features)
# The function uses note_matrix[:, -1] to get the last column (onset times)
# We'll create a small matrix with 5 notes and 4 features
note_matrix = np.array([
    [60, 0.5, 0.8, 0],      # MIDI note, velocity, duration, onset=0
    [62, 0.6, 0.8, 4],      # onset=4
    [64, 0.7, 0.8, 6],      # onset=6
    [65, 0.5, 0.8, 8],      # onset=8
    [67, 0.6, 0.8, 10],     # onset=10
], dtype=np.float32)

# Extract onset column for testing
onset = note_matrix[:, -1].astype(int)

nbpm = 4  # beats per measure
nspb = 4  # sixteenth notes per beat

ARGS = {
    "note_matrix": note_matrix,
    "nbpm": nbpm,
    "nspb": nspb,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (5,) — shape (5,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
chord_bins_to_reduction_mat Not yet run
unverified (harvested, no recorded run)
zzwaang/melody-reduction-algo/tonal_reduction_algo/postprocess.py
code served (permissive licence) · get_code("d061d35354d16c14")
How it ran

no invocation recorded.

chord_id_analysis Not yet run
unverified (harvested, no recorded run)
zzwaang/melody-reduction-algo/tonal_reduction_algo/preprocess.py
code served (permissive licence) · get_code("fff56ef56b160eeb")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 12 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recent deep music generation studies have put much emphasis on long-term generation with structures. However, we are yet to see high-quality, well-structured whole-song generation. In this paper, we make the first attempt to model a full music piece under the realization of compositional hierarchy. With a focus on symbolic representations of pop songs, we define a hierarchical language, in which each level of hierarchy focuses on the semantics and context dependency at a certain music scope. The high-level languages reveal whole-song form, phrase, and cadence, whereas the low-level languages focus on notes, chords, and their local patterns. A cascaded diffusion model is trained to model the hierarchical language, where each level is conditioned on its upper levels. Experiments and analysis show that our model is capable of generating full-piece music with recognizable global verse-chorus structure and cadences, and the music quality is higher than the baselines. Additionally, we show that the proposed model is controllable in a flexible way. By sampling from the interpretable hierarchical languages or adjusting pre-trained external representations, users can control the music flow via various features such as phrase harmonic structures, rhythmic patterns, and accompaniment texture. 1

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2405.09901")
get_code_for_paper("2405.09901")
have("2405.09901")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2405.09901.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2405.09901.svg)](https://syntology.ai/paper/2405.09901)

Connect an agent — have() is free.