1

2026-08-14 13:45:02 +08:00
parent 96348ce39b
commit fba2b60c2a
3 changed files with 712 additions and 0 deletions
@@ -0,0 +1,309 @@
+package main
+
+// Edge TTS 客户端：微软 Edge 朗读私有 WebSocket 协议。
+// 实现对照 edge-tts 7.2.8（drm.py / constants.py / communicate.py）逐行校验：
+// Sec-MS-GEC 纯算法生成（无 HTTP 预请求），403 时读响应 Date 头校正时钟偏移重试。
+
+import (
+	"bytes"
+	"context"
+	"crypto/rand"
+	"crypto/sha256"
+	"encoding/binary"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"math"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/gogf/gf/v2/errors/gerror"
+	"github.com/google/uuid"
+	"github.com/gorilla/websocket"
+)
+
+const (
+	trustedClientToken = "6A5AA1D4EAFF4E9FB37E23D68491D6F4"
+	winEpoch           = 11644473600 // Unix 与 Windows 文件时间纪元的秒差（1601-01-01）
+	originHeader       = "chrome-extension://jdiccldimpdaibmpdkjnbmckianbfold"
+	userAgent          = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36 Edg/143.0.0.0"
+	audioURL           = "wss://speech.platform.bing.com/consumer/speech/synthesize/readaloud/edge/v1"
+	audioFormat        = "audio-24khz-48kbitrate-mono-mp3"
+	maxChunkBytes      = 4096 // 单请求文本字节上限（与 edge-tts 一致，超长分块多轮连接）
+)
+
+type ttsClient struct {
+	voice     string
+	rate      string
+	version   string
+	timeout   time.Duration
+	retries   int
+	clockSkew float64 // 403 时按服务端 Date 头累加校正
+}
+
+// permanentErr 协议类错误，重试无意义（不重试）。
+type permanentErr struct{ err error }
+
+func (e *permanentErr) Error() string { return e.err.Error() }
+func (e *permanentErr) Unwrap() error { return e.err }
+
+// synthesize 合成一段文本，返回 mp3 字节；超长文本按字节上限分块、逐块独立连接后拼接。
+func (c *ttsClient) synthesize(ctx context.Context, text string) ([]byte, error) {
+	text = escapeXML(cleanText(text))
+	var buf []byte
+	for _, chunk := range splitChunks(text, maxChunkBytes) {
+		audio, err := c.synthesizeChunk(ctx, chunk)
+		if err != nil {
+			return nil, err
+		}
+		buf = append(buf, audio...)
+	}
+	if len(buf) == 0 {
+		return nil, gerror.New("合成结果为空")
+	}
+	return buf, nil
+}
+
+func (c *ttsClient) synthesizeChunk(ctx context.Context, text string) ([]byte, error) {
+	var lastErr error
+	for attempt := 0; attempt <= c.retries; attempt++ {
+		if attempt > 0 {
+			select {
+			case <-time.After(time.Duration(attempt) * time.Second):
+			case <-ctx.Done():
+				return nil, ctx.Err()
+			}
+		}
+		audio, err := c.turn(ctx, text)
+		if err == nil {
+			return audio, nil
+		}
+		lastErr = err
+		var pe *permanentErr
+		if errors.As(err, &pe) {
+			return nil, err
+		}
+	}
+	return nil, fmt.Errorf("重试 %d 次仍失败: %w", c.retries, lastErr)
+}
+
+// turn 一次连接完成一帧文本的合成，返回 mp3 字节。
+func (c *ttsClient) turn(ctx context.Context, text string) ([]byte, error) {
+	conn, err := c.dial(ctx)
+	if err != nil {
+		return nil, err
+	}
+	defer conn.Close()
+
+	// 帧 1：speech.config（输出格式配置）
+	configFrame := fmt.Sprintf(
+		"X-Timestamp:%s\r\nContent-Type:application/json; charset=utf-8\r\nPath:speech.config\r\n\r\n"+
+			`{"context":{"synthesis":{"audio":{"metadataoptions":{"sentenceBoundaryEnabled":"true","wordBoundaryEnabled":"false"},"outputFormat":"%s"}}}}`+"\r\n",
+		dateString(), audioFormat)
+	// 帧 2：ssml（xml:lang 固定 en-US，与 edge-tts 一致；X-Timestamp 尾缀 Z 是微软端约定）
+	ssml := fmt.Sprintf(
+		"<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='en-US'>"+
+			"<voice name='%s'><prosody pitch='+0Hz' rate='%s' volume='+0%%'>%s</prosody></voice></speak>",
+		c.voice, c.rate, text)
+	ssmlFrame := fmt.Sprintf(
+		"X-RequestId:%s\r\nContent-Type:application/ssml+xml\r\nX-Timestamp:%sZ\r\nPath:ssml\r\n\r\n%s",
+		strings.ReplaceAll(uuid.NewString(), "-", ""), dateString(), ssml)
+
+	conn.SetWriteDeadline(time.Now().Add(c.timeout))
+	if err := conn.WriteMessage(websocket.TextMessage, []byte(configFrame)); err != nil {
+		return nil, fmt.Errorf("发送 speech.config 失败: %w", err)
+	}
+	if err := conn.WriteMessage(websocket.TextMessage, []byte(ssmlFrame)); err != nil {
+		return nil, fmt.Errorf("发送 ssml 失败: %w", err)
+	}
+
+	var audio []byte
+	for {
+		conn.SetReadDeadline(time.Now().Add(c.timeout))
+		mt, payload, err := conn.ReadMessage()
+		if err != nil {
+			return nil, fmt.Errorf("读帧失败: %w", err)
+		}
+		switch mt {
+		case websocket.TextMessage:
+			path, err := textFramePath(payload)
+			if err != nil {
+				return nil, &permanentErr{err}
+			}
+			switch path {
+			case "turn.end":
+				if len(audio) == 0 {
+					return nil, &permanentErr{gerror.New("无音频返回（文本或参数可能不被支持）")}
+				}
+				return audio, nil
+			case "turn.start", "response", "audio.metadata":
+				// 忽略
+			default:
+				return nil, &permanentErr{fmt.Errorf("未知响应 Path: %s", path)}
+			}
+		case websocket.BinaryMessage:
+			data, err := audioFrameData(payload)
+			if err != nil {
+				return nil, &permanentErr{err}
+			}
+			audio = append(audio, data...)
+		}
+	}
+}
+
+func (c *ttsClient) dial(ctx context.Context) (*websocket.Conn, error) {
+	connID := strings.ReplaceAll(uuid.NewString(), "-", "")
+	wsURL := fmt.Sprintf("%s?TrustedClientToken=%s&ConnectionId=%s&Sec-MS-GEC=%s&Sec-MS-GEC-Version=%s",
+		audioURL, trustedClientToken, connID, c.secMsGec(), c.version)
+	header := http.Header{
+		"Pragma":          {"no-cache"},
+		"Cache-Control":   {"no-cache"},
+		"Origin":          {originHeader},
+		"User-Agent":      {userAgent},
+		"Accept-Language": {"en-US,en;q=0.9"},
+		"Cookie":          {"muid=" + muid() + ";"},
+	}
+	dialer := websocket.Dialer{
+		EnableCompression: true,
+		HandshakeTimeout:  c.timeout,
+	}
+	conn, resp, err := dialer.DialContext(ctx, wsURL, header)
+	if err != nil {
+		if resp != nil && resp.StatusCode == http.StatusForbidden {
+			// 403 多为时钟偏差：读 Date 头校正偏移，下次握手即有效
+			if serverTS, ok := parseRFC2616Date(resp.Header.Get("Date")); ok {
+				c.clockSkew += serverTS - float64(time.Now().Unix())
+			}
+		}
+		return nil, fmt.Errorf("握手失败: %w", err)
+	}
+	return conn, nil
+}
+
+// secMsGec 生成 Sec-MS-GEC 令牌（edge-tts drm.py generate_sec_ms_gec）：
+// (UTC秒 + 时钟偏差 + 11644473600) 取整到 5 分钟窗 → ×1e9/100 转 Windows 100ns 刻度 →
+// 拼接 TrustedClientToken → SHA256 大写 hex。
+func (c *ttsClient) secMsGec() string {
+	ticks := float64(time.Now().Unix()) + c.clockSkew
+	ticks += winEpoch
+	ticks -= math.Mod(ticks, 300)
+	ticks *= 1e9 / 100
+	sum := sha256.Sum256([]byte(fmt.Sprintf("%.0f%s", ticks, trustedClientToken)))
+	return strings.ToUpper(hex.EncodeToString(sum[:]))
+}
+
+// muid 32 位大写 hex 随机串（Cookie: muid=...;）
+func muid() string {
+	b := make([]byte, 16)
+	_, _ = rand.Read(b)
+	return strings.ToUpper(hex.EncodeToString(b))
+}
+
+// dateString JavaScript 风格 UTC 时间串（speech.config / ssml 帧头）
+func dateString() string {
+	return time.Now().UTC().Format("Mon Jan 02 2006 15:04:05 GMT+0000 (Coordinated Universal Time)")
+}
+
+// parseRFC2616Date 解析 HTTP Date 头（如 Thu, 14 Aug 2026 12:34:56 GMT）
+func parseRFC2616Date(s string) (float64, bool) {
+	t, err := time.Parse("Mon, 02 Jan 2006 15:04:05 MST", s)
+	if err != nil {
+		return 0, false
+	}
+	return float64(t.Unix()), true
+}
+
+// cleanText 剔除服务端不支持的 ASCII 控制字符（0-8、11-12、14-31 → 空格）
+func cleanText(s string) string {
+	rs := []rune(s)
+	for i, r := range rs {
+		if (r >= 0 && r <= 8) || (r >= 11 && r <= 12) || (r >= 14 && r <= 31) {
+			rs[i] = ' '
+		}
+	}
+	return string(rs)
+}
+
+// escapeXML 转义 XML 特殊字符（与 saxutils.escape 默认一致：& < >）
+func escapeXML(s string) string {
+	r := strings.NewReplacer("&", "&amp;", "<", "&lt;", ">", "&gt;")
+	return r.Replace(s)
+}
+
+// splitChunks 按字节上限在 rune 边界切分（超长文本分块，避免单请求过大）
+func splitChunks(s string, maxBytes int) []string {
+	if len(s) <= maxBytes {
+		return []string{s}
+	}
+	var chunks []string
+	var buf []byte
+	for _, r := range s {
+		rb := string(r)
+		if len(buf) > 0 && len(buf)+len(rb) > maxBytes {
+			chunks = append(chunks, string(buf))
+			buf = nil
+		}
+		buf = append(buf, rb...)
+	}
+	if len(buf) > 0 {
+		chunks = append(chunks, string(buf))
+	}
+	return chunks
+}
+
+// textFramePath 取文本帧的 Path 头（帧结构：头\r\n\r\n载荷）
+func textFramePath(payload []byte) (string, error) {
+	idx := bytes.Index(payload, []byte("\r\n\r\n"))
+	if idx < 0 {
+		return "", gerror.New("文本帧缺头尾分隔")
+	}
+	return framePath(payload[:idx])
+}
+
+// audioFrameData 解析二进制音频帧：前 2 字节大端头长度，头块（自身含尾部 \r\n）后即 mp3 载荷。
+// 终止帧：无 Content-Type 且载荷为空（实测头文本含尾部 \r\n，与 Python 的 headerLen+2 偏移等价）。
+func audioFrameData(payload []byte) ([]byte, error) {
+	if len(payload) < 2 {
+		return nil, gerror.New("音频帧缺头长度")
+	}
+	headerLen := int(binary.BigEndian.Uint16(payload[:2]))
+	if 2+headerLen > len(payload) {
+		return nil, gerror.New("音频帧头长度越界")
+	}
+	path, err := framePath(payload[2 : 2+headerLen])
+	if err != nil {
+		return nil, err
+	}
+	if path != "audio" {
+		return nil, fmt.Errorf("二进制帧 Path 非 audio: %s", path)
+	}
+	data := payload[2+headerLen:]
+	contentType, _ := frameHeader(payload[2:2+headerLen], "Content-Type")
+	switch {
+	case contentType == "":
+		if len(data) == 0 {
+			return nil, nil // 终止帧
+		}
+		return nil, gerror.New("无 Content-Type 却有数据")
+	case contentType != "audio/mpeg":
+		return nil, fmt.Errorf("意外的 Content-Type: %s", contentType)
+	case len(data) == 0:
+		return nil, gerror.New("audio/mpeg 帧缺音频数据")
+	}
+	return data, nil
+}
+
+func framePath(headers []byte) (string, error) {
+	return frameHeader(headers, "Path")
+}
+
+func frameHeader(headers []byte, key string) (string, error) {
+	for _, line := range bytes.Split(headers, []byte("\r\n")) {
+		k, v, ok := bytes.Cut(line, []byte(":"))
+		if ok && string(k) == key {
+			return string(v), nil
+		}
+	}
+	return "", fmt.Errorf("帧缺 %s 头", key)
+}
@@ -0,0 +1,107 @@
+package main
+
+// gentss：离线生成 TTS 朗读音频（Edge TTS 神经网络语音）并回填各 audio 字段。
+// 用法：
+//
+//	go run ./cmd/gentss --list                        # 打印待生成清单
+//	go run ./cmd/gentss --strategy=1                  # 只生成第 1 计（8 节点剧本已导入）
+//	go run ./cmd/gentss --only=script|content|scene|option|teach   # 只生成指定内容源
+//	go run ./cmd/gentss --force                       # 覆盖已生成 mp3 重新合成
+//
+// 幂等：mp3 文件已存在且字段已回填 → 跳过（--force 覆盖）；文件在但字段空 → 只补回填。
+// 配置在 config.yml 的 tts 段；产物落 ui-src/static/generated/audio/，URL /static/generated/audio/...（与 PNG 素材同挂载）。
+
+import (
+	"flag"
+	"fmt"
+	"sync"
+	"time"
+
+	_ "github.com/gogf/gf/contrib/drivers/sqlite/v2"
+
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/os/glog"
+)
+
+var staticDir = "ui-src/static/generated" // 打包入客户端的音频目录
+
+type genFlags struct {
+	listOnly bool
+	only     string // script | content | scene | option | teach
+	strategy int    // 0 = 全部计策
+	force    bool
+}
+
+type ttsConfig struct {
+	voice       string
+	rate        string
+	version     string
+	concurrency int
+	timeout     int
+	retries     int
+}
+
+func main() {
+	fl := parseFlags()
+	cfg := loadTTSConfig()
+	if fl.listOnly {
+		printInventory(fl)
+		return
+	}
+	client := &ttsClient{
+		voice:   cfg.voice,
+		rate:    cfg.rate,
+		version: cfg.version,
+		timeout: time.Duration(cfg.timeout) * time.Second,
+		retries: cfg.retries,
+	}
+	tasks := enumerateTasks(fl)
+	if len(tasks) == 0 {
+		fmt.Println("[gentss] 无待生成任务")
+		return
+	}
+	sem := make(chan struct{}, cfg.concurrency)
+	var wg sync.WaitGroup
+	for _, t := range tasks {
+		t.force = fl.force
+		wg.Add(1)
+		sem <- struct{}{}
+		go func(t task) {
+			defer wg.Done()
+			defer func() { <-sem }()
+			if err := runTask(client, t); err != nil {
+				glog.Errorf(ctx, "[gentss] %s 失败: %v", taskDesc(t), err)
+			}
+		}(t)
+	}
+	wg.Wait()
+	fmt.Println("[gentss] 完成")
+}
+
+func parseFlags() genFlags {
+	var fl genFlags
+	flag.BoolVar(&fl.listOnly, "list", false, "只打印待生成清单")
+	flag.StringVar(&fl.only, "only", "", "script | content | scene | option | teach，默认全部")
+	flag.IntVar(&fl.strategy, "strategy", 0, "只处理指定计策（strategy_id），0=全部")
+	flag.BoolVar(&fl.force, "force", false, "覆盖已生成音频重新合成")
+	flag.Parse()
+	return fl
+}
+
+func loadTTSConfig() ttsConfig {
+	return ttsConfig{
+		voice:       g.Cfg().MustGet(ctx, "tts.voice", "zh-CN-XiaoxiaoNeural").String(),
+		rate:        g.Cfg().MustGet(ctx, "tts.rate", "+0%").String(),
+		version:     g.Cfg().MustGet(ctx, "tts.version", "1-143.0.3650.75").String(),
+		concurrency: g.Cfg().MustGet(ctx, "tts.concurrency", 2).Int(),
+		timeout:     g.Cfg().MustGet(ctx, "tts.timeout", 60).Int(),
+		retries:     g.Cfg().MustGet(ctx, "tts.retries", 3).Int(),
+	}
+}
+
+func taskDesc(t task) string {
+	if t.kind == "script" {
+		return fmt.Sprintf("node %d", t.nodeID)
+	}
+	return t.relPath
+}
@@ -0,0 +1,296 @@
+package main
+
+// 任务枚举与回填：扫描待合成内容 → 逐任务合成 mp3 → 写 ui-src/static/generated/audio/ → 回填各 audio 字段。
+// 幂等：文件已存在且字段已回填 → 跳过；--force 重生成覆盖；文件在但字段空（半途崩溃）→ 只补回填。
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/gogf/gf/v2/database/gdb"
+	"github.com/gogf/gf/v2/frame/g"
+	"github.com/gogf/gf/v2/os/glog"
+
+	"36wisdom/biz/consts"
+)
+
+var ctx = context.Background()
+
+// task 一条合成任务。kind 决定文件路径与回填字段。
+type task struct {
+	kind       string // script | content | scene | option | teach
+	levelID    int64  // scene/script/content/option 用
+	nodeID     int64  // script/content/option 用
+	optionID   int64
+	strategyID int64 // teach 用
+	text       string
+	relPath    string // staticDir 相对路径，如 audio/1/2/content.mp3
+	scriptJSON string // kind=script 时整列 script JSON
+	force      bool
+}
+
+// scriptLine 台词行（回填 audio 时保持原字段）
+type scriptLine struct {
+	Emotion string `json:"emotion"`
+	Pinyin  string `json:"pinyin"`
+	Speaker string `json:"speaker"`
+	Text    string `json:"text"`
+	Audio   string `json:"audio,omitempty"`
+}
+
+// ---------- 枚举 ----------
+
+func enumerateTasks(fl genFlags) []task {
+	var tasks []task
+	only := map[string]bool{}
+	if fl.only == "" {
+		only["script"], only["content"], only["scene"], only["option"], only["teach"] = true, true, true, true, true
+	} else {
+		only[fl.only] = true
+	}
+	levelIDs, nodeIDs := strategyScope(fl)
+	scopeLevel := func(m *gdb.Model) *gdb.Model {
+		if fl.strategy > 0 {
+			return m.WhereIn("level_id", levelIDs)
+		}
+		return m
+	}
+
+	if only["script"] {
+		m := g.DB().Model(consts.TableSceneNode).Fields("id, level_id, script").Where("script IS NOT NULL AND script != ''")
+		for _, r := range all(scopeLevel(m)) {
+			tasks = append(tasks, task{
+				kind: "script", levelID: r["level_id"].Int64(), nodeID: r["id"].Int64(),
+				scriptJSON: r["script"].String(),
+			})
+		}
+	}
+	if only["content"] {
+		m := g.DB().Model(consts.TableSceneNode).Fields("id, level_id, content").Where("content != '' AND (audio IS NULL OR audio = '')")
+		for _, r := range all(scopeLevel(m)) {
+			nid := r["id"].Int64()
+			tasks = append(tasks, task{
+				kind: "content", levelID: r["level_id"].Int64(), nodeID: nid, text: r["content"].String(),
+				relPath: fmt.Sprintf("audio/%d/%d/content.mp3", r["level_id"].Int64(), nid),
+			})
+		}
+	}
+	if only["scene"] {
+		m := g.DB().Model(consts.TableLevel).Fields("id, scene_content").Where("scene_content != '' AND (scene_audio IS NULL OR scene_audio = '')")
+		if fl.strategy > 0 {
+			m = m.Where("strategy_id", fl.strategy)
+		}
+		for _, r := range all(m) {
+			lid := r["id"].Int64()
+			tasks = append(tasks, task{
+				kind: "scene", levelID: lid, text: r["scene_content"].String(),
+				relPath: fmt.Sprintf("audio/%d/scene.mp3", lid),
+			})
+		}
+	}
+	if only["option"] {
+		m := g.DB().Model(consts.TableNodeOption).Fields("id, node_id, text").Where("text != '' AND (audio IS NULL OR audio = '')")
+		if fl.strategy > 0 {
+			m = m.WhereIn("node_id", nodeIDs)
+		}
+		rows := all(m)
+		levelOf := nodeLevelMap(rows)
+		for _, r := range rows {
+			oid, nid := r["id"].Int64(), r["node_id"].Int64()
+			tasks = append(tasks, task{
+				kind: "option", levelID: levelOf[nid], nodeID: nid, optionID: oid, text: r["text"].String(),
+				relPath: fmt.Sprintf("audio/%d/%d/opt_%d.mp3", levelOf[nid], nid, oid),
+			})
+		}
+	}
+	if only["teach"] {
+		m := g.DB().Model(consts.TableStrategy).Fields("id, teach_content").Where("teach_content != '' AND (teach_audio IS NULL OR teach_audio = '')")
+		if fl.strategy > 0 {
+			m = m.Where("id", fl.strategy)
+		}
+		for _, r := range all(m) {
+			sid := r["id"].Int64()
+			tasks = append(tasks, task{
+				kind: "teach", strategyID: sid, text: r["teach_content"].String(),
+				relPath: fmt.Sprintf("audio/teach/%d.mp3", sid),
+			})
+		}
+	}
+	return tasks
+}
+
+// strategyScope 返回某计策涉及的 level_ids 与 node_ids（strategy<=0 时均 nil）
+func strategyScope(fl genFlags) ([]int64, []int64) {
+	if fl.strategy <= 0 {
+		return nil, nil
+	}
+	var levelIDs []int64
+	for _, v := range arr(g.DB().Model(consts.TableLevel).Where("strategy_id", fl.strategy), "id") {
+		levelIDs = append(levelIDs, v.Int64())
+	}
+	var nodeIDs []int64
+	for _, v := range arr(g.DB().Model(consts.TableSceneNode).WhereIn("level_id", levelIDs), "id") {
+		nodeIDs = append(nodeIDs, v.Int64())
+	}
+	return levelIDs, nodeIDs
+}
+
+// nodeLevelMap 选项行 → 所属关卡（跨表组装：先取 node_ids 再对 scene_node 批量查）
+func nodeLevelMap(optRows []gdb.Record) map[int64]int64 {
+	nodeIDs := make([]int64, 0, len(optRows))
+	for _, r := range optRows {
+		nodeIDs = append(nodeIDs, r["node_id"].Int64())
+	}
+	out := map[int64]int64{}
+	if len(nodeIDs) == 0 {
+		return out
+	}
+	for _, r := range all(g.DB().Model(consts.TableSceneNode).Fields("id, level_id").WhereIn("id", nodeIDs)) {
+		out[r["id"].Int64()] = r["level_id"].Int64()
+	}
+	return out
+}
+
+// ---------- 执行 ----------
+
+func runTask(c *ttsClient, t task) error {
+	if t.kind == "script" {
+		return runScriptTask(c, t)
+	}
+	if strings.TrimSpace(t.text) == "" {
+		return nil
+	}
+	if !t.force && fileExists(filepath.Join(staticDir, t.relPath)) {
+		backfill(t) // 幂等补回填：防上次写文件后未回填
+		return nil
+	}
+	audio, err := c.synthesize(ctx, t.text)
+	if err != nil {
+		return err
+	}
+	if err := writeFile(filepath.Join(staticDir, t.relPath), audio); err != nil {
+		return err
+	}
+	backfill(t)
+	fmt.Printf("[gentss] ok %-7s %s\n", t.kind, t.relPath)
+	return nil
+}
+
+// runScriptTask 整节点台词：逐行合成（无 audio 的行），全部完成后整列重写 script JSON。
+func runScriptTask(c *ttsClient, t task) error {
+	var lines []scriptLine
+	if err := json.Unmarshal([]byte(t.scriptJSON), &lines); err != nil {
+		return fmt.Errorf("script JSON 解析失败: %v", err)
+	}
+	changed := false
+	for i := range lines {
+		if strings.TrimSpace(lines[i].Text) == "" {
+			continue
+		}
+		rel := fmt.Sprintf("audio/%d/%d/%d.mp3", t.levelID, t.nodeID, i)
+		file := filepath.Join(staticDir, rel)
+		if t.force || !fileExists(file) {
+			audio, err := c.synthesize(ctx, lines[i].Text)
+			if err != nil {
+				return fmt.Errorf("第 %d 句合成失败: %w", i, err)
+			}
+			if err := writeFile(file, audio); err != nil {
+				return fmt.Errorf("第 %d 句写文件失败: %w", i, err)
+			}
+		}
+		url := "/static/generated/" + rel
+		if lines[i].Audio != url {
+			lines[i].Audio = url
+			changed = true
+		}
+	}
+	if changed {
+		b, err := json.Marshal(lines)
+		if err != nil {
+			return err
+		}
+		if _, err := g.DB().Model(consts.TableSceneNode).Ctx(ctx).Data(g.Map{"script": string(b)}).Where("id", t.nodeID).Update(); err != nil {
+			return err
+		}
+	}
+	fmt.Printf("[gentss] ok %-7s node %d（%d 句）\n", t.kind, t.nodeID, len(lines))
+	return nil
+}
+
+// backfill 回填 audio 字段（URL 前缀 /static/generated/ 与 PNG 素材同挂载）
+func backfill(t task) {
+	url := "/static/generated/" + t.relPath
+	var err error
+	switch t.kind {
+	case "content":
+		_, err = g.DB().Model(consts.TableSceneNode).Ctx(ctx).Data(g.Map{"audio": url}).Where("id", t.nodeID).Update()
+	case "scene":
+		_, err = g.DB().Model(consts.TableLevel).Ctx(ctx).Data(g.Map{"scene_audio": url}).Where("id", t.levelID).Update()
+	case "option":
+		_, err = g.DB().Model(consts.TableNodeOption).Ctx(ctx).Data(g.Map{"audio": url}).Where("id", t.optionID).Update()
+	case "teach":
+		_, err = g.DB().Model(consts.TableStrategy).Ctx(ctx).Data(g.Map{"teach_audio": url}).Where("id", t.strategyID).Update()
+	}
+	if err != nil {
+		glog.Errorf(ctx, "[gentss] 回填 %s 失败: %v", t.relPath, err)
+	}
+}
+
+// ---------- 清单 ----------
+
+func printInventory(fl genFlags) {
+	tasks := enumerateTasks(fl)
+	if len(tasks) == 0 {
+		fmt.Println("[gentss] 无待生成任务")
+		return
+	}
+	counts := map[string]int{}
+	for _, t := range tasks {
+		counts[t.kind]++
+	}
+	fmt.Printf("待生成任务：script 节点 %d、内容 %d、开场 %d、选项 %d、学堂 %d\n",
+		counts["script"], counts["content"], counts["scene"], counts["option"], counts["teach"])
+	for _, t := range tasks {
+		if t.kind == "script" {
+			fmt.Printf("  script   node %d (level %d)\n", t.nodeID, t.levelID)
+		} else {
+			fmt.Printf("  %-7s %s\n", t.kind, t.relPath)
+		}
+	}
+}
+
+// ---------- 工具 ----------
+
+// all 查询并取全部行（出错即退出，与 genasset 一致）
+func all(m *gdb.Model) gdb.Result {
+	rows, err := m.Ctx(ctx).All()
+	if err != nil {
+		glog.Fatal(ctx, err)
+	}
+	return rows
+}
+
+// arr 查询单列值
+func arr(m *gdb.Model, field string) gdb.Array {
+	vals, err := m.Ctx(ctx).Array(field)
+	if err != nil {
+		glog.Fatal(ctx, err)
+	}
+	return vals
+}
+
+func writeFile(p string, data []byte) error {
+	if err := os.MkdirAll(filepath.Dir(p), 0o755); err != nil {
+		return err
+	}
+	return os.WriteFile(p, data, 0o644)
+}
+
+func fileExists(p string) bool {
+	_, err := os.Stat(p)
+	return err == nil
+}