#!/bin/bash
# ============================================================
# 昔涟语音提取管线
#
# 步骤:
#   1. 从 HSR 音频包提取 .wem (本脚本)
#   2. 用 ww2ogg/vgmstream 转换 .wem → .wav (需手动安装工具)
#   3. 用 ffmpeg 标准化音频格式
# ============================================================
set -e

HSR_AUDIO_DIR="D:/MeowG/Honkai：Star_Rail/StarRail_Data/Persistent/Audio/AudioPackage/Windows/Chinese(PRC)"
RAW_DIR="D:/Project/Code/Uni/Cyrene-Voice-Model/data/raw"
CLEANED_DIR="D:/Project/Code/Uni/Cyrene-Voice-Model/data/cleaned"
SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"

echo "=== 昔涟语音提取管线 ==="
echo ""

# ---- 阶段 1: 从 .pck 提取 .wem ----
echo "[阶段 1/4] 提取候选 .pck 文件..."

# 昔涟是 3.x 角色，语音在以下文件中:
#   - VoBanks 27-31 (最新角色语音库)
#   - External_del_4.0_chapter5_*  (3.0 主线昔涟出场)
#   - External_del_4.1_chapter_*   (3.1 主线昔涟出场)
TARGETS=(
    "VoBanks27.pck"
    "VoBanks28.pck"
    "VoBanks29.pck"
    "VoBanks30.pck"
    "VoBanks31.pck"
    "External_del_4.0_chapter5_0.pck"
    "External_del_4.0_chapter5_1.pck"
    "External_del_4.0_chapter5_2.pck"
    "External_del_4.1_chapter_0.pck"
    "External_del_4.1_chapter_1.pck"
    "External_del_4.1_chapter_2.pck"
)

for target in "${TARGETS[@]}"; do
    pck_path="${HSR_AUDIO_DIR}/${target}"
    if [ -f "$pck_path" ]; then
        echo "  提取: $target ($(du -h "$pck_path" | cut -f1))"
        python3 "${SCRIPT_DIR}/extract_pck.py" "$pck_path" "${RAW_DIR}/${target%.pck}/"
    else
        echo "  跳过: $target (文件不存在)"
    fi
done

echo ""
echo "[阶段 1/4] 完成: .wem 文件已提取到 ${RAW_DIR}/"

# ---- 阶段 2: .wem → .wav 转换 ----
echo ""
echo "[阶段 2/4] 需要转换 .wem → .wav"
echo ""
echo "  HSR 使用 Wwise 专有编码 (0xFFFF)，ffmpeg 无法直接解码。"
echo "  请使用以下工具之一进行转换:"
echo ""
echo "  方案 A — vgmstream CLI (推荐, 最简单):"
echo "    下载: https://github.com/vgmstream/vgmstream/releases"
echo "    解压后将 vgmstream-cli.exe 放入 scripts/voice/tools/"
echo "    然后运行本脚本的 --convert 模式"
echo ""
echo "  方案 B — AnimeWwise GUI (最强大, 保留原始文件名):"
echo "    下载: https://github.com/Escartem/AnimeWwise/releases"
echo "    直接打开 GUI, 选择 HSR 目录, 导出昔涟语音"
echo ""
echo "  方案 C — ww2ogg + revorb (传统方案):"
echo "    下载 ww2ogg.exe + revorb.exe + packed_codebooks.bin"
echo "    放入 scripts/voice/tools/"
echo ""
echo "  安装工具后, 运行: $0 --convert"
echo ""

# ---- 阶段 3 (条件): 批量转换 ----
if [ "$1" = "--convert" ]; then
    echo "[阶段 3/4] 转换 .wem → .wav..."

    TOOLS_DIR="${SCRIPT_DIR}/tools"

    # 优先使用 vgmstream
    if [ -f "${TOOLS_DIR}/vgmstream-cli.exe" ]; then
        echo "  使用 vgmstream-cli..."
        find "${RAW_DIR}" -name "*.wem" | while read wem; do
            wav="${wem%.wem}.wav"
            if [ ! -f "$wav" ]; then
                "${TOOLS_DIR}/vgmstream-cli.exe" -o "$wav" "$wem" 2>/dev/null
            fi
        done
    elif [ -f "${TOOLS_DIR}/ww2ogg.exe" ]; then
        echo "  使用 ww2ogg + ffmpeg..."
        find "${RAW_DIR}" -name "*.wem" | while read wem; do
            ogg="${wem%.wem}.ogg"
            wav="${wem%.wem}.wav"
            if [ ! -f "$wav" ]; then
                "${TOOLS_DIR}/ww2ogg.exe" "$wem" -o "$ogg" --pcb "${TOOLS_DIR}/packed_codebooks.bin" 2>/dev/null
                ffmpeg -y -i "$ogg" -ar 22050 -ac 1 -sample_fmt s16 "$wav" 2>/dev/null
                rm -f "$ogg"
            fi
        done
    else
        echo "  错误: 未找到转换工具, 请先安装 vgmstream 或 ww2ogg"
        exit 1
    fi

    echo "[阶段 3/4] 完成"

    # ---- 阶段 4: 音频标准化 + 分类 ----
    echo ""
    echo "[阶段 4/4] 标准化 + 分类..."

    # 按音频时长初步分类 (语音通常 1-15 秒)
    mkdir -p "${CLEANED_DIR}/daily" "${CLEANED_DIR}/battle" \
             "${CLEANED_DIR}/emotional" "${CLEANED_DIR}/story"

    find "${RAW_DIR}" -name "*.wav" | while read wav; do
        # 获取时长
        duration=$(ffprobe -v quiet -show_entries format=duration \
                   -of default=noprint_wrappers=1:nokey=1 "$wav" 2>/dev/null || echo "0")
        dur_float=$(echo "$duration" | awk '{print int($1 * 1000)}')

        basename=$(basename "$wav" .wav)
        parent=$(basename "$(dirname "$wav")")

        # 分类逻辑
        if [ "$dur_float" -lt 500 ]; then
            # < 0.5s: 可能是战斗短语音 / 语气词
            target_dir="${CLEANED_DIR}/battle"
        elif [ "$dur_float" -gt 15000 ]; then
            # > 15s: 可能是剧情长对话
            target_dir="${CLEANED_DIR}/story"
        elif echo "$parent" | grep -qi "chapter"; then
            target_dir="${CLEANED_DIR}/story"
        elif echo "$parent" | grep -qi "vobanks"; then
            # VoBanks 包含战斗 + 日常语音, 需要人工筛选
            target_dir="${CLEANED_DIR}/daily"
        else
            target_dir="${CLEANED_DIR}/daily"
        fi

        # 用 ffmpeg 标准化: 22.05kHz mono 16bit
        ffmpeg -y -i "$wav" -ar 22050 -ac 1 -sample_fmt s16 \
               "${target_dir}/${parent}_${basename}.wav" 2>/dev/null

        echo -n "."
    done

    echo ""
    echo "[阶段 4/4] 完成: 音频已分类到 ${CLEANED_DIR}/"
    echo ""
    echo "文件分布:"
    echo "  日常对话: $(ls "${CLEANED_DIR}/daily/" 2>/dev/null | wc -l) 个"
    echo "  战斗语音: $(ls "${CLEANED_DIR}/battle/" 2>/dev/null | wc -l) 个"
    echo "  情感表达: $(ls "${CLEANED_DIR}/emotional/" 2>/dev/null | wc -l) 个"
    echo "  剧情对话: $(ls "${CLEANED_DIR}/story/" 2>/dev/null | wc -l) 个"
fi

echo ""
echo "=== 管线完成 ==="