feat(volcengine): 接入 Seed-TTS 2.0 与极速 ASR

constant/volc_speech.go

@@ -0,0 +1,6 @@
+package constant
+
+const (
+	ModelDoubaoSeedTTS20    = "doubao-seed-tts-2.0"
+	ModelDoubaoSeedASRFlash = "doubao-seed-asr-flash"
+)

controller/relay.go

@@ -101,9 +101,11 @@ func Relay(c *gin.Context, relayFormat types.RelayFormat) {
 					"error": newAPIError.ToClaudeError(),
 				})
 			default:
-				c.JSON(newAPIError.StatusCode, gin.H{
-					"error": newAPIError.ToOpenAIError(),
-				})
+				if !c.Writer.Written() {
+					c.JSON(newAPIError.StatusCode, gin.H{
+						"error": newAPIError.ToOpenAIError(),
+					})
+				}
 			}
 		}
 	}()
@@ -384,6 +386,9 @@ func processChannelError(c *gin.Context, channelError types.ChannelError, err *t
 		other["channel_id"] = channelId
 		other["channel_name"] = c.GetString("channel_name")
 		other["channel_type"] = c.GetInt("channel_type")
+		if volcSpeechAudit, exists := c.Get("volc_speech_audit"); exists {
+			other["volc_speech"] = volcSpeechAudit
+		}
 		adminInfo := make(map[string]interface{})
 		adminInfo["use_channel"] = c.GetStringSlice("use_channel")
 		isMultiKey := common.GetContextKeyBool(c, constant.ContextKeyChannelIsMultiKey)

docs/volc-speech.md

@@ -0,0 +1,40 @@
+# 火山语音渠道配置
+
+火山 Seed-TTS 2.0 与录音文件极速 ASR 使用独立的火山语音渠道，公开模型分别为：
+
+- `doubao-seed-tts-2.0`
+- `doubao-seed-asr-flash`
+
+渠道类型选择火山方舟（VolcEngine），但不要与已有文本或图像渠道共用。渠道 Key 支持两种格式：
+
+- 新版控制台：直接填写单段 APP Key，网关使用 `X-Api-Key`。
+- 旧版控制台：填写 `appid|access_token`，网关自动使用旧版鉴权头。
+
+调用方继续使用 OpenAI 兼容接口：
+
+- `POST /v1/audio/speech`
+- `POST /v1/audio/transcriptions`
+
+## TTS 音色
+
+在渠道高级设置中配置 `Seed-TTS 2.0 默认音色 ID`。客户端传 `alloy`、`echo`、`fable`、`onyx`、`nova` 或 `shimmer` 时，网关统一映射到该默认音色；客户端直接传火山 speaker ID 时保持原值。
+
+未配置默认音色时，使用 OpenAI 标准音色的请求会被明确拒绝。协议、资源 ID 和鉴权方式由服务端固定，客户端请求中的 `metadata` 不会覆盖这些字段。
+
+## 计费倍率
+
+两个模型都必须在模型倍率设置中显式配置价格，未配置时不得开放调用。不要在源码中写入火山易变的刊例价格，应按火山控制台当日后付费价格换算：
+
+- TTS 倍率 = `人民币/万字符 ÷ USDExchangeRate ÷ 0.02`
+- ASR 倍率 = `人民币/小时 ÷ USDExchangeRate ÷ 0.12`
+
+用户分组倍率继续作为销售加价系数。
+
+TTS 按火山返回的 `usage.text_words` 结算；仅在成功收到 `SessionFinished` 但缺少 usage 时，回退为 Unicode 字符数。ASR 按每分钟 1000 个内部计费单位预扣，并在完成后使用火山 `audio_info.duration` 校正。
+
+## 支持边界
+
+- TTS 输出：MP3、Opus、PCM；语速范围 0.5–2.0。
+- ASR 输入：WAV、MP3、OGG/Opus；最大 100MB、最长 2 小时。
+- ASR 输出：`json`、`text`、`verbose_json`。
+- 不支持实时 WebSocket ASR、异步长文件 ASR、双向 TTS、SSE、声音复刻和阿里语音。

dto/audio.go

@@ -26,6 +26,10 @@ type AudioRequest struct {
 	XVectorOnlyMode         json.RawMessage `json:"x_vector_only_mode,omitempty"`
 	MaxNewTokens            json.RawMessage `json:"max_new_tokens,omitempty"`
 	InitialCodecChunkFrames json.RawMessage `json:"initial_codec_chunk_frames,omitempty"`
+	// LocalAudioDurationMS 仅保存网关本地解析出的音频时长，不会发送给上游。
+	LocalAudioDurationMS int64 `json:"-"`
+	// LocalAudioFormat 仅保存经过校验的音频格式，不会发送给上游。
+	LocalAudioFormat string `json:"-"`
 	// TODO：ensure that the logic remains correct after the stream is started.
 	//Stream                  json.RawMessage `json:"stream,omitempty"`
 }

dto/channel_settings.go

@@ -53,6 +53,11 @@ type ChannelOtherSettings struct {
 	UpstreamModelUpdateLastRemovedModels  []string              `json:"upstream_model_update_last_removed_models,omitempty"`  // 上次检测到的可删除模型
 	UpstreamModelUpdateIgnoredModels      []string              `json:"upstream_model_update_ignored_models,omitempty"`       // 手动忽略的模型
 	AdvancedCustom                        *AdvancedCustomConfig `json:"advanced_custom,omitempty"`
+	VolcSpeech                            *VolcSpeechConfig     `json:"volc_speech,omitempty"`
+}
+
+type VolcSpeechConfig struct {
+	DefaultTTSSpeaker string `json:"default_tts_speaker,omitempty"`
 }
 
 func (s *ChannelOtherSettings) IsOpenRouterEnterprise() bool {

relay/audio_handler.go

@@ -53,6 +53,18 @@ func AudioHelper(c *gin.Context, info *relaycommon.RelayInfo) (newAPIError *type
 	var httpResp *http.Response
 	if resp != nil {
 		httpResp = resp.(*http.Response)
+		if info.VolcSpeechAudit != nil {
+			info.VolcSpeechAudit.LogID = httpResp.Header.Get("X-Tt-Logid")
+			volcSpeechAudit := map[string]interface{}{
+				"resource_id":   info.VolcSpeechAudit.ResourceID,
+				"protocol":      info.VolcSpeechAudit.Protocol,
+				"billing_units": info.VolcSpeechAudit.BillingUnits,
+			}
+			if info.VolcSpeechAudit.LogID != "" {
+				volcSpeechAudit["log_id"] = info.VolcSpeechAudit.LogID
+			}
+			c.Set("volc_speech_audit", volcSpeechAudit)
+		}
 		if httpResp.StatusCode != http.StatusOK {
 			newAPIError = service.RelayErrorHandler(c.Request.Context(), httpResp, false)
 			// reset status code 重置状态码

relay/channel/volcengine/adaptor.go

@@ -10,6 +10,7 @@ import (
 	"path/filepath"
 	"strings"
 
+	"github.com/QuantumNous/new-api/common"
 	channelconstant "github.com/QuantumNous/new-api/constant"
 	"github.com/QuantumNous/new-api/dto"
 	"github.com/QuantumNous/new-api/relay/channel"
@@ -47,6 +48,51 @@ func (a *Adaptor) ConvertClaudeRequest(c *gin.Context, info *relaycommon.RelayIn
 }
 
 func (a *Adaptor) ConvertAudioRequest(c *gin.Context, info *relaycommon.RelayInfo, request dto.AudioRequest) (io.Reader, error) {
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedTTS20 {
+		if _, _, _, err := parseVolcSpeechCredential(info.ApiKey); err != nil {
+			return nil, err
+		}
+		volcRequest, encoding, err := buildVolcTTSV3Request(request, info.ChannelOtherSettings.VolcSpeech)
+		if err != nil {
+			return nil, err
+		}
+		c.Set(contextKeyResponseFormat, encoding)
+		info.VolcSpeechAudit = &relaycommon.VolcSpeechAuditInfo{
+			ResourceID: volcTTSResourceID,
+			Protocol:   volcTTSProtocol,
+		}
+		setVolcSpeechAuditContext(c, info.VolcSpeechAudit)
+		jsonData, err := common.Marshal(volcRequest)
+		if err != nil {
+			return nil, fmt.Errorf("failed to marshal volcengine TTS request: %w", err)
+		}
+		return bytes.NewReader(jsonData), nil
+	}
+
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedASRFlash {
+		if _, _, _, err := parseVolcSpeechCredential(info.ApiKey); err != nil {
+			return nil, err
+		}
+		if info.RelayMode != constant.RelayModeAudioTranscription {
+			return nil, errors.New("doubao-seed-asr-flash only supports audio transcriptions")
+		}
+		volcRequest, err := buildVolcASRFlashRequest(c, request)
+		if err != nil {
+			return nil, err
+		}
+		c.Set(contextKeyResponseFormat, request.ResponseFormat)
+		info.VolcSpeechAudit = &relaycommon.VolcSpeechAuditInfo{
+			ResourceID: volcASRFlashResourceID,
+			Protocol:   volcASRFlashProtocol,
+		}
+		setVolcSpeechAuditContext(c, info.VolcSpeechAudit)
+		jsonData, err := common.Marshal(volcRequest)
+		if err != nil {
+			return nil, fmt.Errorf("failed to marshal volcengine ASR request: %w", err)
+		}
+		return bytes.NewReader(jsonData), nil
+	}
+
 	if info.RelayMode != constant.RelayModeAudioSpeech {
 		return nil, errors.New("unsupported audio relay mode")
 	}
@@ -237,6 +283,13 @@ func (a *Adaptor) Init(info *relaycommon.RelayInfo) {
 }
 
 func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedTTS20 {
+		return volcTTSV3URL, nil
+	}
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedASRFlash {
+		return volcASRFlashURL, nil
+	}
+
 	baseUrl := info.ChannelBaseUrl
 	if baseUrl == "" {
 		baseUrl = channelconstant.ChannelBaseURLs[channelconstant.ChannelTypeVolcEngine]
@@ -287,6 +340,31 @@ func (a *Adaptor) GetRequestURL(info *relaycommon.RelayInfo) (string, error) {
 func (a *Adaptor) SetupRequestHeader(c *gin.Context, req *http.Header, info *relaycommon.RelayInfo) error {
 	channel.SetupApiRequestHeader(info, c, req)
 
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedTTS20 {
+		headers, err := buildVolcSpeechHeaders(info.ApiKey, volcTTSResourceID, "X-Api-App-Id")
+		if err != nil {
+			return err
+		}
+		for key, values := range headers {
+			(*req)[key] = values
+		}
+		req.Set("X-Control-Require-Usage-Tokens-Return", "*")
+		req.Set("Content-Type", gin.MIMEJSON)
+		return nil
+	}
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedASRFlash {
+		headers, err := buildVolcSpeechHeaders(info.ApiKey, volcASRFlashResourceID, "X-Api-App-Key")
+		if err != nil {
+			return err
+		}
+		for key, values := range headers {
+			(*req)[key] = values
+		}
+		req.Set("X-Api-Sequence", "-1")
+		req.Set("Content-Type", gin.MIMEJSON)
+		return nil
+	}
+
 	if info.RelayMode == constant.RelayModeAudioSpeech {
 		parts := strings.Split(info.ApiKey, "|")
 		if len(parts) == 2 {
@@ -330,6 +408,11 @@ func (a *Adaptor) ConvertOpenAIResponsesRequest(c *gin.Context, info *relaycommo
 }
 
 func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, requestBody io.Reader) (any, error) {
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedTTS20 ||
+		info.OriginModelName == channelconstant.ModelDoubaoSeedASRFlash {
+		return doVolcSpeechRequest(a, c, info, requestBody)
+	}
+
 	if info.RelayMode == constant.RelayModeAudioSpeech {
 		baseUrl := info.ChannelBaseUrl
 		if baseUrl == "" {
@@ -346,6 +429,13 @@ func (a *Adaptor) DoRequest(c *gin.Context, info *relaycommon.RelayInfo, request
 }
 
 func (a *Adaptor) DoResponse(c *gin.Context, resp *http.Response, info *relaycommon.RelayInfo) (usage any, err *types.NewAPIError) {
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedTTS20 {
+		return handleVolcTTSV3Response(c, resp, info, c.GetString(contextKeyResponseFormat))
+	}
+	if info.OriginModelName == channelconstant.ModelDoubaoSeedASRFlash {
+		return handleVolcASRFlashResponse(c, resp, info, c.GetString(contextKeyResponseFormat))
+	}
+
 	if info.RelayFormat == types.RelayFormatClaude {
 		if _, ok := channelconstant.ChannelSpecialBases[info.ChannelBaseUrl]; ok {
 			adaptor := claude.Adaptor{}