data/freeze_test_spatial_metadata.py

#!/usr/bin/env python3
"""
Freeze spatial rendering metadata into test JSON files for deterministic evaluation.

This script pre-computes and stores the spatial parameters (SOFA file, mic positions,
source positions, HRTF indices) into each test JSON so that every evaluation run
produces identical binaural rendering.

Usage:
    conda activate semhear_emma2
    python data/freeze_test_spatial_metadata.py --mixtures_dir data/audio_mixtures_old --hrtf_dir data/hrtf
"""

import argparse
import glob
import hashlib
import json
import os
import random
import sys

import numpy as np

# Add project root to path so we can import data.multi_ch_simulator
project_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.insert(0, project_root)

from data.multi_ch_simulator import Multi_Ch_Simulator


def freeze_spatial_metadata(mixtures_dir: str, hrtf_dir: str, sr: int = 44100,
                            reverb: bool = True, dry_run: bool = False) -> None:
    """
    For each test JSON in mixtures_dir/test/, compute deterministic spatial
    rendering params and write them into the JSON file.
    """
    test_dir = os.path.join(mixtures_dir, "test")
    json_files = sorted(glob.glob(os.path.join(test_dir, "*.json")))

    # Filter out manifest files
    json_files = [f for f in json_files if not os.path.basename(f).startswith("_")]

    if not json_files:
        print(f"No JSON files found in {test_dir}")
        return

    print(f"Found {len(json_files)} test JSON files in {test_dir}")

    # Initialize the multi-channel simulator (same as dataloader)
    simulator_pool = Multi_Ch_Simulator(hrtf_dir, "test", sr, reverb)

    updated = 0
    skipped = 0

    for json_path in json_files:
        with open(json_path, "r") as f:
            metadata = json.load(f)

        # Build the list of spatial labels (same logic as dataloader)
        spatial_labels = ["speech"]
        for distractor_name in metadata.get("distractors", []):
            spatial_labels.append(distractor_name)
        num_sources = len(spatial_labels)

        # Compute deterministic seed from audio file path (same as dataloader)
        audio_file = json_path.replace(".json", ".wav")
        seed = int.from_bytes(
            hashlib.sha256(str(audio_file).encode()).digest()[:4], "little"
        )

        # Seed BEFORE simulator selection — this is the critical fix
        np.random.seed(seed)
        random.seed(seed)

        # Select simulator (now deterministic)
        sim = simulator_pool.get_random_simulator()

        # Generate spatial params (also deterministic with the seed set)
        sim.initialize_room_with_random_params(
            num_sources, 0, spatial_labels, nbackground_sources=0
        )

        # Extract spatial metadata
        spatial_meta = sim.get_metadata()

        # Convert numpy types to native Python for JSON serialization
        spatial_meta = _make_json_serializable(spatial_meta)

        # Write spatial fields into the metadata
        metadata["sofa"] = spatial_meta["sofa"]
        metadata["mic_positions"] = spatial_meta["mic_positions"]
        metadata["sources"] = spatial_meta["sources"]
        metadata["num_background"] = spatial_meta["num_background"]
        metadata["duration"] = spatial_meta["duration"]

        if dry_run:
            print(f"  [dry-run] {os.path.basename(json_path)}: "
                  f"sofa={metadata['sofa']}, "
                  f"sources={len(metadata['sources'])}")
            skipped += 1
        else:
            with open(json_path, "w") as f:
                json.dump(metadata, f, indent=2)
            updated += 1

        if (updated + skipped) % 500 == 0:
            print(f"  Processed {updated + skipped}/{len(json_files)} files...")

    print(f"Done: {updated} updated, {skipped} skipped (dry_run={dry_run})")


def _make_json_serializable(obj):
    """Recursively convert numpy types to native Python types."""
    if isinstance(obj, dict):
        return {k: _make_json_serializable(v) for k, v in obj.items()}
    elif isinstance(obj, (list, tuple)):
        return [_make_json_serializable(item) for item in obj]
    elif isinstance(obj, np.integer):
        return int(obj)
    elif isinstance(obj, np.floating):
        return float(obj)
    elif isinstance(obj, np.ndarray):
        return obj.tolist()
    return obj


def main():
    parser = argparse.ArgumentParser(
        description="Freeze spatial rendering metadata into test JSON files"
    )
    parser.add_argument(
        "--mixtures_dir", required=True,
        help="Path to audio_mixtures directory (contains test/ subdir)"
    )
    parser.add_argument(
        "--hrtf_dir", default="data/hrtf",
        help="Path to HRTF directory (default: data/hrtf)"
    )
    parser.add_argument(
        "--sr", type=int, default=44100,
        help="Sample rate (default: 44100)"
    )
    parser.add_argument(
        "--no-reverb", action="store_true",
        help="Disable reverb (use CIPIC only)"
    )
    parser.add_argument(
        "--dry-run", action="store_true",
        help="Print what would be written without modifying files"
    )
    args = parser.parse_args()

    freeze_spatial_metadata(
        mixtures_dir=args.mixtures_dir,
        hrtf_dir=args.hrtf_dir,
        sr=args.sr,
        reverb=not args.no_reverb,
        dry_run=args.dry_run,
    )


if __name__ == "__main__":
    main()