export.ps1

# 确保SOFA环境已激活！
# 如果你还没激活，请先运行:
# conda activate SOFA

# --- 定义变量方便管理路径 ---
$SOFAPython = "$env:CONDA_PREFIX\python.exe" # 使用SOFA环境的python路径
$BuildScript = "D:\DiffSingerDatasets\MakeDiffSinger\acoustic_forced_alignment\build_dataset.py"
$AddScript = "D:\DiffSingerDatasets\MakeDiffSinger\variance-temp-solution\add_ph_num.py" # 修正后的路径
$InferScript = "D:\DiffSingerDatasets\MakeDiffSinger\SOME\batch_infer.py"
$Dictionary = "D:\DiffSingerDatasets\SOFA\dictionary\opencpop-extension.txt"
$ModelCheckpoint = "D:\DiffSingerDatasets\MakeDiffSinger\SOME\pretrained\0119_continuous256_5spk\model_ckpt_steps_100000_simplified.ckpt"

$SourceDataDir = "D:\DiffSingerDatasets\255_step2_man"
$OutputDatasetDir = "D:\DiffSingerDatasets\255_built_man"
$SourceWavDir = Join-Path $SourceDataDir "wav"
$SourceTgDir = Join-Path $SourceDataDir "TextGrid"
$OutputCsvPath = Join-Path $OutputDatasetDir "transcriptions.csv"

Write-Host "--- 开始处理数据集: $($SourceDataDir) ---"

# --- 步骤 1: 构建基础数据集结构 ---
Write-Host "`n--- 运行 build_dataset.py ---"
# 确保输出目录存在
mkdir $OutputDatasetDir -Force

& $SOFAPython $BuildScript --wavs $SourceWavDir --tg $SourceTgDir --dataset $OutputDatasetDir

# 检查 build_dataset.py 是否成功 (简单检查CSV是否存在)
if (-not (Test-Path $OutputCsvPath)) {
    Write-Host "!!! 错误: build_dataset.py 未能生成 $OutputCsvPath 文件。请检查上一步的输出。终止后续步骤。" -ForegroundColor Red
    exit 1 # 退出脚本
}

# --- 步骤 2: 添加 ph_num 到 CSV ---
Write-Host "`n--- 运行 add_ph_num.py ---"
# add_ph_num.py 期望CSV路径作为第一个参数
& $SOFAPython $AddScript $OutputCsvPath --dictionary $Dictionary

# Note: add_ph_num.py 通常不返回错误码，或者错误处理在内部。这里假设它会成功或打印警告。
# 如果需要严格检查，可以解析其标准输出来判断是否成功添加了 ph_num 列。

# --- 步骤 3: 自动音高推断 ---
Write-Host "`n--- 运行 batch_infer.py (音高推断) ---"
# 检查模型文件是否存在
if (-not (Test-Path $ModelCheckpoint)) {
    Write-Host "!!! 错误: 模型文件未找到: $($ModelCheckpoint)。跳过音高推断步骤。" -ForegroundColor Red
} else {
    & $SOFAPython $InferScript --model $ModelCheckpoint --dataset $OutputDatasetDir --overwrite
}

Write-Host "`n--- 数据集处理完成: $($OutputDatasetDir) ---"