build_test_csv.py

import os
import csv
import re

REC_ROOT = "audio/recorded"
CLEAN_ROOT = "audio/clean"
ASR_DIR = "annotations/asr_label"
VAD_DIR = "annotations/speaker_activity"

rows = []

for location in sorted(os.listdir(REC_ROOT)):
    loc_dir = os.path.join(REC_ROOT, location)
    if not os.path.isdir(loc_dir):
        continue

    for wav in sorted(os.listdir(loc_dir)):
        if not wav.endswith(".wav"):
            continue

        # 例:
        # doa_0-240-120_session-001_ovlp_75-100_spk0-3570_spk1-1188_spk2-8463.wav

        # 提取 session_key
        m = re.match(r"^(doa_[^_]+)_(session-.*)\.wav$", wav)
        if not m:
            continue

        session_key = m.group(2)
        recorded_path = f"{REC_ROOT}/{location}/{wav}"

        # 找该 session 的所有 speaker clean 文件
        for spk_id in ["spk0", "spk1", "spk2"]:
            clean_filename = f"{session_key}_{spk_id}_clean.wav"
            clean_path = f"{CLEAN_ROOT}/{clean_filename}"

            if not os.path.exists(clean_path):
                continue

            asr_filename = f"{session_key}_{spk_id}_label.txt"
            asr_path = f"{ASR_DIR}/{asr_filename}"

            vad_filename = f"{session_key}_{spk_id}_clean.txt"
            vad_path = f"{VAD_DIR}/{vad_filename}"

            if not os.path.exists(asr_path):
                continue

            sample_id = f"{location}_{session_key}_{spk_id}"

            rows.append([
                sample_id,
                location,
                session_key,
                spk_id,
                recorded_path,
                clean_path,
                asr_path,
                vad_path
            ])

with open("test.csv", "w", newline="") as f:
    writer = csv.writer(f)
    writer.writerow([
        "id",
        "location",
        "session_key",
        "speaker_id",
        "recorded_path",
        "clean_path",
        "asr_label_path",
        "speaker_activity_path"
    ])
    writer.writerows(rows)

print("Generated test.csv with", len(rows), "rows")