fix: 兼容上游 sentence 字段为对象/字符串两种形态,加 debug 打印无 event 元数据行

adapter/volcano/response.go

@@ -72,7 +72,7 @@ func ParseStream(body io.Reader, started time.Time) (*ParsedStream, error) {
 		switch resp.Event {
 		case "TTSSentenceStart":
 			if common.DebugLog {
-				log.Printf("Sentence start: sequence=%d, sentence=%s", resp.Sequence, resp.Sentence)
+				log.Printf("Sentence start: sequence=%d, sentence=%s", resp.Sequence, resp.SentenceText())
 			}
 		case "TTSSentenceEnd":
 			if common.DebugLog {
@@ -81,7 +81,7 @@ func ParseStream(body io.Reader, started time.Time) (*ParsedStream, error) {
 		case "TTSSubtitle":
 			if resp.Data != "" {
 				out.Subtitles = append(out.Subtitles, dto.SubtitleEntry{
-					Text:     resp.Sentence,
+					Text:     resp.SentenceText(),
 					Sequence: resp.Sequence,
 				})
 			}
@@ -105,10 +105,15 @@ func ParseStream(body io.Reader, started time.Time) (*ParsedStream, error) {
 				gotFirstChunk = true
 			}
 		case "":
-			// 传输帧,跳过
+			// 无 event 字段的元数据行(上游 TTSSentenceStart 偶发省略 event),
+			// debug 模式打印便于排查;正常情况静默跳过。
+			if common.DebugLog && resp.Code == 0 {
+				log.Printf("volcano: 无 event 字段 code=%d sequence=%d sentence=%s data_len=%d", resp.Code, resp.Sequence, resp.SentenceText(), len(resp.Data))
+			}
+		}
 		default:
 			if common.DebugLog {
-				log.Printf("volcano: 忽略未识别事件 event=%q sequence=%d", resp.Event, resp.Sequence)
+				log.Printf("volcano: 忽略未识别事件 event=%q sequence=%d sentence=%s data_len=%d", resp.Event, resp.Sequence, resp.SentenceText(), len(resp.Data))
 			}
 		}
 	}

dto/tts.go

@@ -1,6 +1,9 @@
 package dto
 
-import "time"
+import (
+	"encoding/json"
+	"time"
+)
 
 // OpenAITTSRequest 是 /v1/audio/speech 接收的请求体。
 // 仅 input / speed / response_format 实际影响火山侧;
@@ -14,16 +17,40 @@ type OpenAITTSRequest struct {
 }
 
 // V3TTSResponse 是火山 v3 HTTP Chunked 流式响应中每一行的 JSON 结构。
+// Sentence 字段上游有时返回字符串(TTSSentenceStart 里的句文本),有时返回对象
+// ({"phonemes":[...],"text":"...","words":[...]}),用 json.RawMessage 兼容两种形态,
+// 避免任意一种上游变更都导致整行解析失败。
 type V3TTSResponse struct {
-	ReqID    string   `json:"reqid"`
+	ReqID    string          `json:"reqid"`
-	Code     int      `json:"code"`
+	Code     int             `json:"code"`
-	Message  string   `json:"message"`
+	Message  string          `json:"message"`
-	Event    string   `json:"event"`
+	Event    string          `json:"event"`
-	Sequence int      `json:"sequence"`
+	Sequence int             `json:"sequence"`
-	Data     string   `json:"data"`
+	Data     string          `json:"data"`
-	Sentence string   `json:"sentence,omitempty"`
+	Sentence json.RawMessage `json:"sentence,omitempty"`
-	IsFinal  bool     `json:"is_final"`
+	IsFinal  bool            `json:"is_final"`
-	Usage    *V3Usage `json:"usage,omitempty"`
+	Usage    *V3Usage        `json:"usage,omitempty"`
+}
+
+// SentenceText 从 Sentence 提取可读文本:
+//   - 字符串直接返回
+//   - 对象尝试取 .text 字段
+//   - 其它情况返回原始 JSON
+func (r *V3TTSResponse) SentenceText() string {
+	if len(r.Sentence) == 0 {
+		return ""
+	}
+	var s string
+	if err := json.Unmarshal(r.Sentence, &s); err == nil {
+		return s
+	}
+	var obj struct {
+		Text string `json:"text"`
+	}
+	if err := json.Unmarshal(r.Sentence, &obj); err == nil && obj.Text != "" {
+		return obj.Text
+	}
+	return string(r.Sentence)
 }
 
 // V3Usage 由 X-Control-Require-Usage-Tokens-Return 触发,包含计费字符数。