Heiheihaha17's picture
Add files using upload-large-folder tool
347d9a3 verified
Raw History Blame Contribute Delete
2.76 kB
"""Shared base peak scaling, with restart-safe journals and power provenance."""
from __future__ import annotations
import copy
from pathlib import Path
import numpy as np
import soundfile as sf
from .common import *
def normalize_base(root,index):
ensure_compute()
report_path=root/'reports/bases'/f'{index:04d}.json';report=read(report_path)
desired=.9/max(r['raw_peak'] for r in report['recordings'])
journal_path=root/'private/normalization'/f'{index:04d}.json'
journal=read(journal_path) if journal_path.exists() else dict(gain=desired,files={})
if journal['gain']!=desired:raise ValueError('Normalization gain changed')
if journal.get('complete'):
for record in report['recordings']:
if sha(root/record['path'])!=record['sha256']:raise ValueError('Normalized audio changed')
return report
previous=report['common_final_gain'];ratio=desired/previous
for record in report['recordings']:
path=root/record['path'];key=record['recording_id'];current=sha(path)
entry=journal['files'].get(key)
if entry and current==entry['after_sha256']:
record.update(sha256=current,common_final_gain=desired,final_peak=entry['peak'])
continue
if current!=(entry['before_sha256'] if entry else record['sha256']):raise ValueError('Unexpected pre-normalization audio')
wave,rate=sf.read(path,dtype='float32');wave=(wave*ratio).astype(np.float32)
peak=float(np.max(np.abs(wave)))
if rate!=SR or not np.isfinite(wave).all() or peak>.900001:raise ValueError('Normalization overflow')
temporary=path.with_suffix('.normalized.tmp.wav');write_float_wav(temporary,wave)
entry=dict(before_sha256=current,after_sha256=sha(temporary),peak=peak)
journal['files'][key]=entry;write(journal_path,journal)
temporary.replace(path);record.update(sha256=entry['after_sha256'],common_final_gain=desired,final_peak=peak)
for measurement in report['acoustic_measurements']:
for key in ['target_power','competing_speech_power','environmental_noise_power']:
raw=measurement.get('raw_'+key,measurement[key])
measurement['raw_'+key]=raw;measurement[key]=raw*desired**2
measurement['power_stage']='published FOA after common base peak normalization'
report.update(common_final_gain=desired,final_normalization_applied=True,
peak_normalization=dict(target=.9,scope='same gain across all robots and all three base derivatives',code_sha256=sha(Path(__file__))))
write(report_path,report);journal['complete']=True;write(journal_path,journal)
return report
def normalize(root):
for p in sorted((root/'reports/bases').glob('*.json')):normalize_base(root,int(p.stem))