@xdfnet/ispeak 1.6.1 → 1.6.3

package/Docs/ARCHITECTURE.md

@@ -34,7 +34,7 @@ iSpeak 是一个运行在 macOS 上的本地 TTS 播报守护进程，通过 Uni
 │             ▼                                               │
 │  Speak Worker (single)                                      │
 │  - pending_synth -> speaking                                │
-│  - 调用 TTS 流式接口（失败重试1次）                          │
+│  - 调用 TTS 流式接口（失败直接删除，不重试）                  │
 │  - SSE audio chunk -> StreamPlayer.Write                    │
 │  - 播放完成后删除任务；连续失败删除任务                       │
 │                                                             │
@@ -107,12 +107,14 @@ pending_synth -> speaking -> delete
 
 `Submit(cleanedText, voice, cfg)` 原子执行：
 1. 删除所有 `pending_synth` 任务
-2. 创建新任务（`pending_synth`）
-3. 唤醒 speak worker
+2. 打断当前 `speaking` 任务（取消合成/停止播放）
+3. 创建新任务（`pending_synth`）
+4. 唤醒 speak worker
 
 策略说明：
-- 只清理“未开始合成”的任务
-- 不打断 `speaking`
+- 未开始合成的旧任务直接删除
+- 已领取但过期的旧任务在合成前跳过
+- 正在合成/播放的旧任务会被新消息取消
 
 ### Speak worker 规则
 
@@ -130,9 +132,22 @@ pending_synth -> speaking -> delete
 `handleConnection()`：
 - 读取 socket 文本
 - 解析 `{source:xxx}` 音色前缀
-- `cleanText()` 过滤 Markdown/表格符号
+- `cleanText()` 生成语音友好的文本
 - 将“过滤后文本”提交给 `TaskEngine.Submit`
 
+`cleanText()` 只影响 TTS 播报，不改变屏幕显示内容。当前清洗规则：
+
+- Markdown 格式符号：标题、加粗、反引号、引用符
+- Markdown 表格整块：表头、分隔线、表格内容
+- 代码块、artifact、HTML 页面源码
+- Markdown 链接 URL，仅保留链接标题
+- 绝对路径简化为“路径”
+- 长 commit hash、UUID、长 ID
+- 明显文件列表、模型分片列表、下载清单
+- 下载进度、速度、进度条、ANSI 控制符等终端噪声
+
+清洗目标是保留适合听的内容：结论、成功/失败状态、下一步动作、关键错误原因。
+
 ### 2. 流式合成播放阶段
 
 - speak worker 领取任务
@@ -151,9 +166,9 @@ pending_synth -> speaking -> delete
 
 ## 失败与成本策略
 
-- 新任务到达时仅清理 `pending_synth`，避免无效合成
-- 流式合成/播放失败：整条播报重试 1 次后删除
-- 执行中任务不打断，行为稳定、可预期
+- 新任务到达时清理 `pending_synth` 并打断当前任务，避免无效合成/播放
+- 流式合成/播放失败：直接删除任务，不重试，避免重复播报
+- 只保留最新消息优先播报，降低 TTS 成本
 
 ## 文件布局
 

package/README.md

@@ -1,6 +1,6 @@
 # iSpeak
 
-![Version](https://img.shields.io/badge/version-1.6.1-blue)
+![Version](https://img.shields.io/badge/version-1.6.3-blue)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![Go Version](https://img.shields.io/badge/Go-1.26-blue)](https://golang.org/dl/)
 ![Platform](https://img.shields.io/badge/platform-macOS-green)
@@ -51,11 +51,12 @@ brew install ffmpeg
 git clone https://github.com/xdfnet/iSpeak.git && cd iSpeak && make install
 ```
 
-安装时手动输入 API Key，然后验证：
+安装后编辑 API Key，然后验证：
 
 ```bash
+open ~/.config/iSpeak/config.json
 ispeak status
-ispeak test "iSpeak 准备好了"
+ispeak "iSpeak 准备好了"
 ```
 
 ## 工作原理
@@ -88,6 +89,26 @@ ispeak test "iSpeak 准备好了"
 pending_synth → speaking → delete
 ```
 
+新消息到达时会清理未开始任务，并打断当前合成/播放，只保留最新消息优先播报。
+
+## 语音清洗规则
+
+清洗只影响 TTS 播报内容，不改变 Claude/Codex 屏幕显示内容。
+
+播报前会过滤或简化这些内容：
+
+- Markdown 格式符号：标题 `#`、加粗 `**`、反引号、引用 `>`
+- Markdown 表格整块：表头、分隔线、表格内容都不播
+- 代码块：``` 包裹的内容不播
+- artifact / HTML 内容：不播生成的页面源码
+- Markdown 链接：只保留链接标题，不播 URL
+- 绝对路径：简化为“路径”
+- 长 commit hash、UUID、长 ID：不播
+- 明显文件列表：如模型分片、代码文件列表、下载文件清单
+- 下载进度和终端噪声：百分比、速度、进度条、ANSI 控制符
+
+保留优先级：结论、成功/失败状态、需要用户操作的下一步、关键错误原因。
+
 ## 全部命令
 
 ```bash

package/configs/hook-speak.sh

@@ -11,52 +11,107 @@ LOG="$HOME/.config/iSpeak/hook.log"
 
 input=$(cat)
 
-# 从 stdin JSON 提取 transcript 路径和最后一条消息
-# 简单 JSON 解析（不依赖 python3）
-transcript=$(echo "$input" | sed -n 's/.*"transcript_path"\s*:\s*"\([^"]*\)".*/\1/p')
-last_msg=$(echo "$input" | sed -n 's/.*"last_assistant_message"\s*:\s*"\([^"]*\)".*/\1/p')
+json_value() {
+  local key="$1"
+  if command -v node >/dev/null 2>&1; then
+    printf "%s" "$input" | node -e '
+      const key = process.argv[1];
+      let input = "";
+      process.stdin.setEncoding("utf8");
+      process.stdin.on("data", chunk => input += chunk);
+      process.stdin.on("end", () => {
+        try {
+          const value = JSON.parse(input)[key];
+          if (typeof value === "string") process.stdout.write(value);
+        } catch (_) {}
+      });
+    ' "$key"
+    return
+  fi
 
-all_text="$last_msg"
+  printf "%s" "$input" | sed -n "s/.*\"$key\"[[:space:]]*:[[:space:]]*\"\([^\"]*\)\".*/\1/p"
+}
 
-# 如果有 transcript 文件，提取最近 30 秒内的所有 assistant 消息
-if [[ -n "$transcript" && -f "$transcript" ]]; then
-  # 计算 30 秒前的时间戳（毫秒）
-  cutoff=$(($(date +%s) * 1000 - 30000))
+extract_recent_assistant_text() {
+  local transcript="$1"
+  local cutoff="$2"
 
-  # 用 awk 解析 JSON lines，提取 role=assistant 且 timestamp > cutoff 的 text
-  extra=$(awk -v cutoff="$cutoff" '
+  if command -v node >/dev/null 2>&1; then
+    node -e '
+      const fs = require("fs");
+      const file = process.argv[1];
+      const cutoff = Number(process.argv[2]);
+      const out = [];
+
+      function collectText(content) {
+        if (typeof content === "string") {
+          out.push(content);
+          return;
+        }
+        if (!Array.isArray(content)) return;
+        for (const item of content) {
+          if (item && typeof item.text === "string") out.push(item.text);
+        }
+      }
+
+      for (const line of fs.readFileSync(file, "utf8").split(/\r?\n/)) {
+        if (!line.trim()) continue;
+        try {
+          const event = JSON.parse(line);
+          if (typeof event.timestamp === "number" && event.timestamp < cutoff) continue;
+          if (event.role === "assistant") collectText(event.content);
+          if (event.message && event.message.role === "assistant") collectText(event.message.content);
+        } catch (_) {}
+      }
+      process.stdout.write([...new Set(out.filter(Boolean))].join(" "));
+    ' "$transcript" "$cutoff" 2>/dev/null
+    return
+  fi
+
+  awk -v cutoff="$cutoff" '
     {
-      # 提取 timestamp
-      if (match($0, /"timestamp"\s*:\s*[0-9]+/)) {
+      if (match($0, /"timestamp"[[:space:]]*:[[:space:]]*[0-9]+/)) {
         ts = substr($0, RSTART, RLENGTH)
         gsub(/[^0-9]/, "", ts)
         ts = int(ts)
         if (ts < cutoff) next
       }
 
-      # 提取 role
-      if (match($0, /"role"\s*:\s*"assistant"/)) {
-        # 提取 content（可能是字符串或数组）
-        if (match($0, /"content"\s*:\s*\[/)) {
-          # 数组形式，提取所有 text 字段
+      if (match($0, /"role"[[:space:]]*:[[:space:]]*"assistant"/)) {
+        if (match($0, /"content"[[:space:]]*:[[:space:]]*\[/)) {
           gsub(/[^{]*\[/, "", $0)
           gsub(/\].*/, "", $0)
-          while (match($0, /"text"\s*:\s*"[^"]*"/)) {
+          while (match($0, /"text"[[:space:]]*:[[:space:]]*"[^"]*"/)) {
             t = substr($0, RSTART, RLENGTH)
-            gsub(/"text"\s*:\s*"/, "", t)
+            gsub(/"text"[[:space:]]*:[[:space:]]*"/, "", t)
             gsub(/"$/, "", t)
             if (t != "") print t
             $0 = substr($0, RSTART + RLENGTH)
           }
-        } else if (match($0, /"content"\s*:\s*"\([^"]*\)"/)) {
+        } else if (match($0, /"content"[[:space:]]*:[[:space:]]*"[^"]*"/)) {
           t = substr($0, RSTART, RLENGTH)
-          gsub(/"content"\s*:\s*"/, "", t)
+          gsub(/"content"[[:space:]]*:[[:space:]]*"/, "", t)
           gsub(/"$/, "", t)
           if (t != "") print t
         }
       }
     }
-  ' "$transcript" 2>/dev/null | sort -u | tr '\n' ' ')
+  ' "$transcript" 2>/dev/null | sort -u | tr '\n' ' '
+}
+
+# 从 stdin JSON 提取 transcript 路径和最后一条消息
+transcript=$(json_value "transcript_path")
+last_msg=$(json_value "last_assistant_message")
+
+all_text="$last_msg"
+
+# 如果有 transcript 文件，提取最近 30 秒内的所有 assistant 消息
+if [[ -n "$transcript" && -f "$transcript" ]]; then
+  # 计算 30 秒前的时间戳（毫秒）
+  cutoff=$(($(date +%s) * 1000 - 30000))
+
+  # 优先用 JSON parser，Node 不存在时回退到简易 awk。
+  extra=$(extract_recent_assistant_text "$transcript" "$cutoff")
 
   if [[ -n "$extra" ]]; then
     all_text="$extra"

package/main.go

@@ -17,6 +17,7 @@ import (
 	"os/exec"
 	"os/signal"
 	"path/filepath"
+	"regexp"
 	"strings"
 	"sync"
 	"syscall"
@@ -25,11 +26,6 @@ import (
 	"gopkg.in/natefinch/lumberjack.v2"
 )
 
-const (
-	ttsMaxAttempts  = 2
-	ttsRetryBackoff = 400 * time.Millisecond
-)
-
 var configDir = os.ExpandEnv("$HOME/.config/iSpeak")
 
 var (
@@ -47,6 +43,25 @@ var tempDir string
 
 var errAlreadyRunning = errors.New("iSpeak already running")
 
+var (
+	markdownLinkRe     = regexp.MustCompile(`\[[^\]]+\]\(([^)]*)\)`)
+	absolutePathRe     = regexp.MustCompile(`/(?:Users|private|tmp|var|opt|usr|bin|sbin|etc|Library|Applications)/\S+`)
+	commitHashRe       = regexp.MustCompile(`\b[0-9a-f]{7,40}\b`)
+	uuidRe             = regexp.MustCompile(`\b[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}\b`)
+	urlRe              = regexp.MustCompile(`https?://\S+`)
+	ansiEscapeRe       = regexp.MustCompile(`\x1b\[[0-9;]*[A-Za-z]`)
+	multiSpaceRe       = regexp.MustCompile(`\s+`)
+	markdownListRe     = regexp.MustCompile(`^\s*(?:[-*+]\s+|\d+[.)]\s+)`)
+	htmlTagRe          = regexp.MustCompile(`<[^>]+>`)
+	codeFenceStartRe   = regexp.MustCompile("^```")
+	artifactStartRe    = regexp.MustCompile(`(?i)^<artifact\b`)
+	htmlDocumentLineRe = regexp.MustCompile(`(?i)^<!doctype html|^<html\b|^<head\b|^<body\b|^<style\b|^</`)
+	progressNoiseRe    = regexp.MustCompile(`(?i)(^\s*\d{1,3}%\s*$|\d{1,3}%.*\d+(?:\.\d+)?\s*(?:kb|mb|gb)/s|\bETA\b|^\s*[-=]{3,}\s*$)`)
+	speedNoiseRe       = regexp.MustCompile(`(?i)\d+(?:\.\d+)?\s*(?:kb|mb|gb)/s`)
+	etaNoiseRe         = regexp.MustCompile(`(?i)\bETA\b|预计剩余|剩余时间`)
+	fileListNoiseRe    = regexp.MustCompile(`(?i)\.(?:go|js|ts|tsx|jsx|json|md|yaml|yml|toml|sum|mod|lock|html|css|sh|plist|safetensors|mp3|wav|png|jpg|jpeg|pdf|docx)\b`)
+)
+
 type StreamPlayer interface {
 	Write(audio []byte) error
 	CloseAndWait() error
@@ -54,9 +69,13 @@ type StreamPlayer interface {
 }
 
 type ffplayStreamPlayer struct {
-	path  string
-	cmd   *exec.Cmd
-	stdin io.WriteCloser
+	path string
+	cmd  *exec.Cmd
+
+	mu       sync.Mutex
+	stdin    io.WriteCloser
+	waitOnce sync.Once
+	waitErr  error
 }
 
 func newDefaultStreamPlayer() (StreamPlayer, error) {
@@ -99,35 +118,55 @@ func (p *ffplayStreamPlayer) Write(audio []byte) error {
 	if len(audio) == 0 {
 		return nil
 	}
-	if _, err := p.stdin.Write(audio); err != nil {
+	p.mu.Lock()
+	stdin := p.stdin
+	p.mu.Unlock()
+	if stdin == nil {
+		return fmt.Errorf("播放器输入已关闭")
+	}
+	if _, err := stdin.Write(audio); err != nil {
 		return fmt.Errorf("写入播放器失败: %w", err)
 	}
 	return nil
 }
 
 func (p *ffplayStreamPlayer) CloseAndWait() error {
-	if p.stdin != nil {
-		if err := p.stdin.Close(); err != nil {
+	p.mu.Lock()
+	stdin := p.stdin
+	p.stdin = nil
+	p.mu.Unlock()
+	if stdin != nil {
+		if err := stdin.Close(); err != nil {
 			return fmt.Errorf("关闭播放器输入失败: %w", err)
 		}
-		p.stdin = nil
 	}
-	if err := p.cmd.Wait(); err != nil {
+	if err := p.wait(); err != nil {
 		return fmt.Errorf("ffplay failed: %w", err)
 	}
 	return nil
 }
 
 func (p *ffplayStreamPlayer) Abort() error {
-	if p.stdin != nil {
-		_ = p.stdin.Close()
-		p.stdin = nil
+	p.mu.Lock()
+	stdin := p.stdin
+	p.stdin = nil
+	p.mu.Unlock()
+	if stdin != nil {
+		_ = stdin.Close()
 	}
 	if p.cmd != nil && p.cmd.Process != nil {
 		_ = p.cmd.Process.Kill()
-		_ = p.cmd.Wait()
 	}
-	return nil
+	return p.wait()
+}
+
+func (p *ffplayStreamPlayer) wait() error {
+	p.waitOnce.Do(func() {
+		if p.cmd != nil {
+			p.waitErr = p.cmd.Wait()
+		}
+	})
+	return p.waitErr
 }
 
 type bufferedStreamPlayer struct {
@@ -183,8 +222,11 @@ type TaskEngine struct {
 	mu sync.Mutex
 
 	nextID       uint64
+	latestID     uint64
 	tasks        map[uint64]*Task
 	pendingSynth []uint64
+	activeID     uint64
+	activeCancel context.CancelFunc
 
 	synthWake chan struct{}
 
@@ -207,7 +249,6 @@ func (e *TaskEngine) Start() {
 
 func (e *TaskEngine) Submit(text string, voice VoiceInfo, cfg Config) uint64 {
 	e.mu.Lock()
-	defer e.mu.Unlock()
 
 	// 新任务进来先删所有未开始合成任务
 	for _, id := range e.pendingSynth {
@@ -216,6 +257,12 @@ func (e *TaskEngine) Submit(text string, voice VoiceInfo, cfg Config) uint64 {
 	}
 	e.pendingSynth = e.pendingSynth[:0]
 
+	cancelActive := e.activeCancel
+	activeID := e.activeID
+	if activeID != 0 {
+		log.Printf("打断当前播报任务: id=%d", activeID)
+	}
+
 	e.nextID++
 	task := &Task{
 		ID:     e.nextID,
@@ -225,10 +272,16 @@ func (e *TaskEngine) Submit(text string, voice VoiceInfo, cfg Config) uint64 {
 		Cfg:    cfg,
 	}
 	e.tasks[task.ID] = task
+	e.latestID = task.ID
 	e.pendingSynth = append(e.pendingSynth, task.ID)
 	log.Printf("任务创建: id=%d text=%s", task.ID, text)
 
 	notify(e.synthWake)
+	e.mu.Unlock()
+
+	if cancelActive != nil {
+		cancelActive()
+	}
 	return task.ID
 }
 
@@ -252,24 +305,27 @@ func (e *TaskEngine) processSpeakTask(id uint64) {
 		}
 	}()
 
+	ctx, cancel := context.WithCancel(context.Background())
+	e.setActiveTask(id, cancel)
+	defer e.clearActiveTask(id)
+
 	task, ok := e.getTask(id)
 	if !ok {
 		return
 	}
-
-	var lastErr error
-	for i := 1; i <= ttsMaxAttempts; i++ {
-		lastErr = e.speakOnce(context.Background(), task)
-		if lastErr == nil {
-			break
-		}
-		if i < ttsMaxAttempts {
-			time.Sleep(ttsRetryBackoff)
-		}
+	if !e.isLatestTask(id) {
+		cancel()
+		log.Printf("跳过过期播报任务: id=%d", id)
+		e.deleteTask(id)
+		return
 	}
 
-	if lastErr != nil {
-		log.Printf("播报失败并删除任务: id=%d err=%v", id, lastErr)
+	if err := e.speakOnce(ctx, task); err != nil {
+		if errors.Is(err, context.Canceled) {
+			log.Printf("播报已打断并删除任务: id=%d", id)
+		} else {
+			log.Printf("播报失败并删除任务: id=%d err=%v", id, err)
+		}
 		e.deleteTask(id)
 		return
 	}
@@ -299,9 +355,21 @@ func (e *TaskEngine) speakOnce(ctx context.Context, task *Task) error {
 		return err
 	}
 	log.Printf("TTS 流结束: id=%d elapsed=%s", task.ID, time.Since(startedAt).Round(time.Millisecond))
-	if err := player.CloseAndWait(); err != nil {
+
+	done := make(chan error, 1)
+	go func() {
+		done <- player.CloseAndWait()
+	}()
+	select {
+	case err := <-done:
+		if err != nil {
+			_ = player.Abort()
+			return err
+		}
+	case <-ctx.Done():
 		_ = player.Abort()
-		return err
+		<-done
+		return ctx.Err()
 	}
 	return nil
 }
@@ -344,6 +412,28 @@ func (e *TaskEngine) deleteTask(id uint64) {
 	delete(e.tasks, id)
 }
 
+func (e *TaskEngine) setActiveTask(id uint64, cancel context.CancelFunc) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	e.activeID = id
+	e.activeCancel = cancel
+}
+
+func (e *TaskEngine) clearActiveTask(id uint64) {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	if e.activeID == id {
+		e.activeID = 0
+		e.activeCancel = nil
+	}
+}
+
+func (e *TaskEngine) isLatestTask(id uint64) bool {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.latestID == id
+}
+
 func notify(ch chan struct{}) {
 	select {
 	case ch <- struct{}{}:
@@ -387,6 +477,17 @@ func loadConfig() Config {
 		}
 		var cfg Config
 		if json.Unmarshal(data, &cfg) == nil && cfg.APIKey != "" {
+			if err := validateConfig(cfg); err != nil {
+				log.Printf("配置文件无效: %s err=%v", p, err)
+				configCacheMu.Lock()
+				if configCacheValid {
+					cached := configCache
+					configCacheMu.Unlock()
+					return cached
+				}
+				configCacheMu.Unlock()
+				return cfg
+			}
 			log.Printf("配置文件: %s", p)
 			if cfg.DefaultVoice != nil {
 				log.Printf("默认音色: %s (%s)", cfg.DefaultVoice.VoiceType, cfg.DefaultVoice.ResourceID)
@@ -637,29 +738,56 @@ func extractAudioBase64(event map[string]any) string {
 	return ""
 }
 
-// 过滤格式符号，保留自然朗读文本
+// 过滤格式符号，保留自然朗读文本。
+// 顺序很重要：先跳过跨行块结构，再跳过整行噪声，最后清理行内符号。
 func cleanText(text string) string {
 	var lines []string
-	for _, line := range strings.Split(text, "\n") {
+	rawLines := strings.Split(text, "\n")
+	inCodeBlock := false
+	inArtifact := false
+	inMarkdownTable := false
+	for i := 0; i < len(rawLines); i++ {
+		line := rawLines[i]
 		line = strings.TrimSpace(line)
-		if strings.HasPrefix(line, "|---") || strings.HasPrefix(line, "|:---") {
+		if line == "" {
+			inMarkdownTable = false
+			continue
+		}
+		if codeFenceStartRe.MatchString(line) {
+			inCodeBlock = !inCodeBlock
+			continue
+		}
+		if inCodeBlock {
+			continue
+		}
+		if artifactStartRe.MatchString(line) {
+			inArtifact = !strings.Contains(strings.ToLower(line), "</artifact>")
 			continue
 		}
-		if strings.HasPrefix(line, "---") && strings.Count(line, "-") > 3 {
+		if inArtifact {
+			if strings.Contains(strings.ToLower(line), "</artifact>") {
+				inArtifact = false
+			}
+			continue
+		}
+		if isMarkdownTableSeparator(line) {
+			if len(lines) > 0 && isMarkdownTableRow(strings.TrimSpace(rawLines[i-1])) {
+				lines = lines[:len(lines)-1]
+			}
+			inMarkdownTable = true
 			continue
 		}
-		// 过滤纯表格分隔行（|---|---|、:---|:---| 等）
-		if strings.Trim(line, "|-: ") == "" {
+		if inMarkdownTable {
+			if isMarkdownTableRow(line) {
+				continue
+			}
+			inMarkdownTable = false
+		}
+		if shouldSkipSpeechLine(line) {
 			continue
 		}
-		cleaned := strings.NewReplacer(
-			"**", "",
-			"*", "",
-			"`", "",
-			"#", "",
-			">", "",
-		).Replace(line)
-		cleaned = strings.TrimSpace(cleaned)
+
+		cleaned := cleanSpeechLine(line)
 		if cleaned != "" {
 			lines = append(lines, cleaned)
 		}
@@ -667,6 +795,110 @@ func cleanText(text string) string {
 	return strings.Join(lines, "，")
 }
 
+func shouldSkipSpeechLine(line string) bool {
+	if isMarkdownTableSeparator(line) {
+		return true
+	}
+	if strings.HasPrefix(line, "---") && strings.Count(line, "-") > 3 {
+		return true
+	}
+	if htmlDocumentLineRe.MatchString(line) {
+		return true
+	}
+	if isProgressNoiseLine(line) {
+		return true
+	}
+	if isMostlyTableRow(line) {
+		return true
+	}
+	if isMostlyFileListLine(line) {
+		return true
+	}
+	return false
+}
+
+func isMarkdownTableSeparator(line string) bool {
+	line = strings.TrimSpace(line)
+	return strings.Contains(line, "|") && strings.Trim(line, "|-: ") == ""
+}
+
+func isMarkdownTableRow(line string) bool {
+	line = strings.TrimSpace(line)
+	return strings.Count(line, "|") >= 2
+}
+
+func cleanSpeechLine(line string) string {
+	// Markdown 链接必须在 URL 删除前处理，否则会丢掉链接标题。
+	line = ansiEscapeRe.ReplaceAllString(line, "")
+	line = markdownListRe.ReplaceAllString(line, "")
+	line = markdownLinkRe.ReplaceAllStringFunc(line, func(match string) string {
+		if end := strings.Index(match, "]"); end > 1 {
+			return match[1:end]
+		}
+		return ""
+	})
+	line = urlRe.ReplaceAllString(line, "")
+	line = absolutePathRe.ReplaceAllString(line, " 路径 ")
+	// UUID 必须在短 hash 前处理，避免先删短片段后破坏 UUID 识别。
+	line = uuidRe.ReplaceAllString(line, "")
+	line = commitHashRe.ReplaceAllString(line, "")
+	line = htmlTagRe.ReplaceAllString(line, "")
+	line = strings.NewReplacer(
+		"**", "",
+		"*", "",
+		"`", "",
+		"#", "",
+		">", "",
+		"✅", "",
+		"❌", "",
+		"✓", "",
+		"✗", "",
+		"→", "到",
+	).Replace(line)
+	line = strings.Trim(line, " \t-:|")
+	line = multiSpaceRe.ReplaceAllString(line, " ")
+	return strings.TrimSpace(line)
+}
+
+func isMostlyTableRow(line string) bool {
+	if !strings.Contains(line, "|") {
+		return false
+	}
+	return strings.Count(line, "|") >= 2 && len([]rune(line)) > 40
+}
+
+func isProgressNoiseLine(line string) bool {
+	if !progressNoiseRe.MatchString(line) {
+		return false
+	}
+	if speedNoiseRe.MatchString(line) || etaNoiseRe.MatchString(line) {
+		return true
+	}
+	return !containsCJK(line)
+}
+
+func isMostlyFileListLine(line string) bool {
+	if !fileListNoiseRe.MatchString(line) {
+		return false
+	}
+	if containsCJK(line) {
+		return false
+	}
+	if strings.Contains(line, ".safetensors") {
+		return true
+	}
+	return strings.Count(line, ".") >= 2 || strings.Contains(line, "/") || strings.Contains(line, " - ")
+}
+
+func containsCJK(s string) bool {
+	for _, r := range s {
+		if r >= '\u4e00' && r <= '\u9fff' {
+			return true
+		}
+	}
+	return false
+}
+
 func main() {
 	log.SetFlags(log.Ltime | log.Lshortfile)
 
@@ -779,8 +1011,26 @@ func validateConfig(cfg Config) error {
 	if cfg.Endpoint == "" {
 		return fmt.Errorf("endpoint 未设置")
 	}
-	if cfg.DefaultVoice == nil || cfg.DefaultVoice.VoiceType == "" {
-		return fmt.Errorf("defaultVoice 未设置")
+	if err := validateVoiceInfo("defaultVoice", cfg.DefaultVoice); err != nil {
+		return err
+	}
+	for source, voice := range cfg.SourceVoices {
+		if err := validateVoiceInfo(fmt.Sprintf("sourceVoices.%s", source), voice); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func validateVoiceInfo(name string, voice *VoiceInfo) error {
+	if voice == nil {
+		return fmt.Errorf("%s 未设置", name)
+	}
+	if voice.VoiceType == "" {
+		return fmt.Errorf("%s.voice_type 未设置", name)
+	}
+	if voice.ResourceID == "" {
+		return fmt.Errorf("%s.resourceId 未设置", name)
 	}
 	return nil
 }
@@ -809,13 +1059,24 @@ func handleConnection(conn net.Conn, engine *TaskEngine) {
 	defer conn.Close()
 
 	cfg := loadConfig()
+	if err := validateConfig(cfg); err != nil {
+		log.Printf("配置错误，跳过本次播报: %v", err)
+		return
+	}
 
 	var sb strings.Builder
 	scanner := bufio.NewScanner(conn)
 	scanner.Buffer(make([]byte, 1*1024*1024), 1*1024*1024)
 	for scanner.Scan() {
+		if sb.Len() > 0 {
+			sb.WriteByte('\n')
+		}
 		sb.WriteString(scanner.Text())
 	}
+	if err := scanner.Err(); err != nil {
+		log.Printf("读取 socket 消息失败: %v", err)
+		return
+	}
 
 	text := strings.TrimSpace(sb.String())
 	if text == "" {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xdfnet/ispeak",
-  "version": "1.6.1",
+  "version": "1.6.3",
   "description": "Local macOS TTS daemon for AI coding assistants, powered by Volcengine streaming TTS.",
   "license": "MIT",
   "homepage": "https://github.com/xdfnet/iSpeak#readme",

package/scripts/ispeak

@@ -2,7 +2,7 @@
 # ispeak — iSpeak 控制命令
 set -euo pipefail
 
-VERSION="1.6.1"
+VERSION="1.6.3"
 SOCK="$HOME/.config/iSpeak/ispeak.sock"
 PLIST="$HOME/Library/LaunchAgents/com.iSpeak.plist"
 CMD_NAME="$(basename "$0")"