Heiheihaha17's picture
Add files using upload-large-folder tool
2e20433 verified
Raw History Blame Contribute Delete
7.16 kB
"""Freeze one static parent's room and longer, aligned real-speech excerpts.
Author-side preparation. Public reproduction starts at the bundled recipe and
source FLACs and does not require the author's static manifests or local paths.
"""
from __future__ import annotations
import argparse
import copy
import gzip
import json
import shutil
from pathlib import Path
import numpy as np
import soundfile as sf
from .common import SR, VERSION, position, sha, speech_activity, uid, write
from .motion import SHIFTS, revised_recipe
def rows(path):
with gzip.open(path, 'rt') as f:
for line in f:
yield json.loads(line)
def prepare(release, corpus, out):
parent_id = 'b44440882fb683045c9bafb5'
parent = next(r for r in rows(release/'unified_test.jsonl.gz') if r['scene_id']==parent_id)
room = next(r for r in rows(release/'rooms.jsonl.gz') if r['room_id']==parent['room_id'] and r['split']=='test')
out.mkdir(parents=True, exist_ok=True)
assets = out/'assets/room'
assets.mkdir(parents=True, exist_ok=True)
from generation.native.build_static_v1_scene import build_glb
surfaces = [(s['name'], s['class_'], s['location'], s['size']) for s in room['layout']['surfaces']]
build_glb(assets/'geometry.glb', surfaces, room['room_id'])
shutil.copyfile(assets/'geometry.glb', assets/'geometry.semantic.glb')
for name, value in room['files'].items():
(assets/name).write_text(value)
for name, h in room['asset_hashes'].items():
if sha(assets/name)!=h:
raise ValueError('Room asset failed pinned checksum: '+name)
write(assets/'layout.json', room['layout'])
materials = {m['scene_source_id']:m for m in parent['materials']}
alignments = {}
hashes = {m['sha256'] for m in materials.values()}
for row in rows(release/'alignment_crops.jsonl.gz'):
c = row['crop']
if c['source_sha256'] in hashes:
alignments.setdefault(c['source_sha256'], c['words'])
source_dir = out/'assets/sources'
source_dir.mkdir(parents=True, exist_ok=True)
specs = {'H01':[(1,8,.35),(8,13,4.5)], 'H02':[(0,11,1.5),(11,16,5.1)], 'R02':[(3,7,3.1)]}
physical = {s['source_id']:s for s in parent['physical_scene']['sources']}
entities = []
for sid in ('H01','H02','R02'):
m = materials[sid]
source_path = corpus/'extracted'/m['corpus_path']
if sha(source_path)!=m['sha256']:
raise ValueError('Original source mismatch')
target = source_dir/(sid+'.flac')
shutil.copyfile(source_path, target)
wave, sr = sf.read(target, dtype='float32')
if sr!=SR or wave.ndim!=1:
raise ValueError('Expected original 16 kHz mono speech')
events=[]
for index,(first,last,time_s) in enumerate(specs[sid]):
ws=alignments[m['sha256']][first:last]
a=max(0,round((ws[0]['start_s']-.03)*SR))
b=min(len(wave),round((ws[-1]['end_s']+.04)*SR))
start=round(time_s*SR)
events.append(dict(event_id=uid('event',sid,index), source_id=sid,
audio_asset='assets/sources/'+sid+'.flac', original_source_sha256=m['sha256'],
original_corpus_path=m['corpus_path'], crop_samples=[a,b], start_sample=start,
end_sample=start+b-a, text=' '.join(w['word'] for w in ws),
activity_samples=speech_activity(wave[a:b],start), words=[dict(word=w['word'],
start_sample=start+round(w['start_s']*SR)-a,
end_sample=start+round(w['end_s']*SR)-a, alignment_score=w['mean_score']) for w in ws]))
entities.append(dict(entity_id=sid, entity_type=m['role'], speaker_id=m['speaker_id'],
initial_xyz_m=physical[sid]['position_m'], yaw_deg=0., displacement_m=[0.,0.,0.],
motion_duration_s=6.3, emitted_rms_dbfs=-22. if sid!='R02' else -25., events=events))
robots=[]
for r in parent['physical_scene']['robot_poses']:
robots.append(dict(entity_id=r['robot_id'],entity_type='robot',initial_xyz_m=r['microphone_position_m'],
yaw_deg=r['yaw_deg'],displacement_m=[0.,0.,0.],motion_duration_s=6.3,
playback_role='silent' if r['robot_id']=='R01' else 'playback',
coincident_microphone_and_loudspeaker_centers=True))
shifts=SHIFTS
scenes=[]
for branch,moving in [('easy_human',['H01']),('easy_robot',['R01']),('hard_joint',list(shifts))]:
s=dict(schema_version=VERSION,scene_id=uid(parent_id,branch),parent_static_scene_id=parent_id,
scene_group_id=uid(parent_id,'paired_motion'),branch=branch,
difficulty='hard' if branch=='hard_joint' else 'easy',room_id=room['room_id'],
duration_samples=parent['frames'],sample_rate_hz=SR,evaluation_step_samples=1600,
moving_entity_ids=moving,humans=copy.deepcopy(entities[:2]),robots=copy.deepcopy(robots),
playback_sources=copy.deepcopy(entities[2:]),strict_sync=True,
own_playback_policy='ideal_own_only',coordinate_convention=dict(world_up='+Y',
azimuth_positive='left',yaw_zero='world -Z',yaw_positive='toward world +X'))
for e in s['humans']+s['robots']:
if e['entity_id'] in moving:
e['displacement_m']=shifts[e['entity_id']]
for e in s['playback_sources']:
r=next(r for r in s['robots'] if r['entity_id']==e['entity_id'])
e['displacement_m']=r['displacement_m'];e['yaw_deg']=r['yaw_deg']
scenes.append(s)
settings=copy.deepcopy(parent['renderer']);settings['threadCount']=2
recipe=dict(schema_version=VERSION,seed=20261001,parent_static_scene_id=parent_id,
parent_static_duration_samples=parent['frames'],parent_release_sha256=sha(release/'release.json'),
speech_style='aligned real LibriSpeech excerpts; assembled turns, not natural conversation',
source_license='CC BY 4.0',room=room['layout'],room_assets_sha256=room['asset_hashes'],
foa_mapping=room['mapping'],renderer=settings,source_node_step_s=.2,receiver_node_step_s=.3,
rendering_method='two-time source-emission / receiver-reception RIR basis interpolation; quasi-static propagation',
motion_model='minimum-jerk horizontal translation; fixed height and yaw',scenes=scenes)
write(out/'data/recipe.json',revised_recipe(recipe))
# The checked-in long-utterance alignment selects the current diagnostic demo.
if (out/'assets/sources/H01_alignment.json').exists():
selection=json.loads((out/'assets/sources/H01_alignment.json').read_text())
shutil.copyfile(corpus/'extracted'/selection['original_corpus_path'],out/'assets/sources/H01.flac')
from .continuous import prepare as continuous_prepare
continuous_prepare(out)
print('Prepared',len(scenes),'scenes; duration',parent['frames']/SR,'s',flush=True)
if __name__=='__main__':
p=argparse.ArgumentParser();p.add_argument('--release',type=Path,required=True)
p.add_argument('--corpus',type=Path,required=True);p.add_argument('--out',type=Path,required=True)
a=p.parse_args();prepare(a.release,a.corpus,a.out)