feat: gemini audio input billing

relay/channel/gemini/dto.go

@@ -116,6 +116,12 @@ type GeminiUsageMetadata struct {
 	CandidatesTokenCount int                         `json:"candidatesTokenCount"`
 	CandidatesTokenCount int                         `json:"candidatesTokenCount"`
 	TotalTokenCount      int                         `json:"totalTokenCount"`
 	TotalTokenCount      int                         `json:"totalTokenCount"`
 	ThoughtsTokenCount   int                         `json:"thoughtsTokenCount"`
 	ThoughtsTokenCount   int                         `json:"thoughtsTokenCount"`
+	PromptTokensDetails  []GeminiPromptTokensDetails `json:"promptTokensDetails"`
+}
+
+type GeminiPromptTokensDetails struct {
+	Modality   string `json:"modality"`
+	TokenCount int    `json:"tokenCount"`
 }
 }
 
 
 // Imagen related structs
 // Imagen related structs

relay/channel/gemini/relay-gemini-native.go

@@ -55,6 +55,16 @@ func GeminiTextGenerationHandler(c *gin.Context, resp *http.Response, info *rela
 		TotalTokens:      geminiResponse.UsageMetadata.TotalTokenCount,
 		TotalTokens:      geminiResponse.UsageMetadata.TotalTokenCount,
 	}
 	}
 
 
+	usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
+
+	for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
+		if detail.Modality == "AUDIO" {
+			usage.PromptTokensDetails.AudioTokens = detail.TokenCount
+		} else if detail.Modality == "TEXT" {
+			usage.PromptTokensDetails.TextTokens = detail.TokenCount
+		}
+	}
+
 	// 直接返回 Gemini 原生格式的 JSON 响应
 	// 直接返回 Gemini 原生格式的 JSON 响应
 	jsonResponse, err := json.Marshal(geminiResponse)
 	jsonResponse, err := json.Marshal(geminiResponse)
 	if err != nil {
 	if err != nil {
@@ -100,6 +110,14 @@ func GeminiTextGenerationStreamHandler(c *gin.Context, resp *http.Response, info
 			usage.PromptTokens = geminiResponse.UsageMetadata.PromptTokenCount
 			usage.PromptTokens = geminiResponse.UsageMetadata.PromptTokenCount
 			usage.CompletionTokens = geminiResponse.UsageMetadata.CandidatesTokenCount
 			usage.CompletionTokens = geminiResponse.UsageMetadata.CandidatesTokenCount
 			usage.TotalTokens = geminiResponse.UsageMetadata.TotalTokenCount
 			usage.TotalTokens = geminiResponse.UsageMetadata.TotalTokenCount
+			usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
+			for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
+				if detail.Modality == "AUDIO" {
+					usage.PromptTokensDetails.AudioTokens = detail.TokenCount
+				} else if detail.Modality == "TEXT" {
+					usage.PromptTokensDetails.TextTokens = detail.TokenCount
+				}
+			}
 		}
 		}
 
 
 		// 直接发送 GeminiChatResponse 响应
 		// 直接发送 GeminiChatResponse 响应
@@ -118,7 +136,6 @@ func GeminiTextGenerationStreamHandler(c *gin.Context, resp *http.Response, info
 	}
 	}
 
 
 	// 计算最终使用量
 	// 计算最终使用量
-	usage.PromptTokensDetails.TextTokens = usage.PromptTokens
 	usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
 	usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
 
 
 	// 移除流式响应结尾的[Done]，因为Gemini API没有发送Done的行为
 	// 移除流式响应结尾的[Done]，因为Gemini API没有发送Done的行为

relay/channel/gemini/relay-gemini.go

@@ -313,13 +313,13 @@ func CovertGemini2OpenAI(textRequest dto.GeneralOpenAIRequest, info *relaycommon
 				if part.GetInputAudio().Data == "" {
 				if part.GetInputAudio().Data == "" {
 					return nil, fmt.Errorf("only base64 audio is supported in gemini")
 					return nil, fmt.Errorf("only base64 audio is supported in gemini")
 				}
 				}
-				format, base64String, err := service.DecodeBase64FileData(part.GetInputAudio().Data)
+				base64String, err := service.DecodeBase64AudioData(part.GetInputAudio().Data)
 				if err != nil {
 				if err != nil {
 					return nil, fmt.Errorf("decode base64 audio data failed: %s", err.Error())
 					return nil, fmt.Errorf("decode base64 audio data failed: %s", err.Error())
 				}
 				}
 				parts = append(parts, GeminiPart{
 				parts = append(parts, GeminiPart{
 					InlineData: &GeminiInlineData{
 					InlineData: &GeminiInlineData{
-						MimeType: format,
+						MimeType: "audio/" + part.GetInputAudio().Format,
 						Data:     base64String,
 						Data:     base64String,
 					},
 					},
 				})
 				})
@@ -771,6 +771,13 @@ func GeminiChatStreamHandler(c *gin.Context, resp *http.Response, info *relaycom
 			usage.CompletionTokens = geminiResponse.UsageMetadata.CandidatesTokenCount
 			usage.CompletionTokens = geminiResponse.UsageMetadata.CandidatesTokenCount
 			usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
 			usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
 			usage.TotalTokens = geminiResponse.UsageMetadata.TotalTokenCount
 			usage.TotalTokens = geminiResponse.UsageMetadata.TotalTokenCount
+			for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
+				if detail.Modality == "AUDIO" {
+					usage.PromptTokensDetails.AudioTokens = detail.TokenCount
+				} else if detail.Modality == "TEXT" {
+					usage.PromptTokensDetails.TextTokens = detail.TokenCount
+				}
+			}
 		}
 		}
 		err = helper.ObjectData(c, response)
 		err = helper.ObjectData(c, response)
 		if err != nil {
 		if err != nil {
@@ -845,6 +852,14 @@ func GeminiChatHandler(c *gin.Context, resp *http.Response, info *relaycommon.Re
 	usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
 	usage.CompletionTokenDetails.ReasoningTokens = geminiResponse.UsageMetadata.ThoughtsTokenCount
 	usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
 	usage.CompletionTokens = usage.TotalTokens - usage.PromptTokens
 
 
+	for _, detail := range geminiResponse.UsageMetadata.PromptTokensDetails {
+		if detail.Modality == "AUDIO" {
+			usage.PromptTokensDetails.AudioTokens = detail.TokenCount
+		} else if detail.Modality == "TEXT" {
+			usage.PromptTokensDetails.TextTokens = detail.TokenCount
+		}
+	}
+
 	fullTextResponse.Usage = usage
 	fullTextResponse.Usage = usage
 	jsonResponse, err := json.Marshal(fullTextResponse)
 	jsonResponse, err := json.Marshal(fullTextResponse)
 	if err != nil {
 	if err != nil {

relay/relay-text.go

@@ -352,6 +352,7 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	promptTokens := usage.PromptTokens
 	promptTokens := usage.PromptTokens
 	cacheTokens := usage.PromptTokensDetails.CachedTokens
 	cacheTokens := usage.PromptTokensDetails.CachedTokens
 	imageTokens := usage.PromptTokensDetails.ImageTokens
 	imageTokens := usage.PromptTokensDetails.ImageTokens
+	audioTokens := usage.PromptTokensDetails.AudioTokens
 	completionTokens := usage.CompletionTokens
 	completionTokens := usage.CompletionTokens
 	modelName := relayInfo.OriginModelName
 	modelName := relayInfo.OriginModelName
 
 
@@ -367,6 +368,7 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	dPromptTokens := decimal.NewFromInt(int64(promptTokens))
 	dPromptTokens := decimal.NewFromInt(int64(promptTokens))
 	dCacheTokens := decimal.NewFromInt(int64(cacheTokens))
 	dCacheTokens := decimal.NewFromInt(int64(cacheTokens))
 	dImageTokens := decimal.NewFromInt(int64(imageTokens))
 	dImageTokens := decimal.NewFromInt(int64(imageTokens))
+	dAudioTokens := decimal.NewFromInt(int64(audioTokens))
 	dCompletionTokens := decimal.NewFromInt(int64(completionTokens))
 	dCompletionTokens := decimal.NewFromInt(int64(completionTokens))
 	dCompletionRatio := decimal.NewFromFloat(completionRatio)
 	dCompletionRatio := decimal.NewFromFloat(completionRatio)
 	dCacheRatio := decimal.NewFromFloat(cacheRatio)
 	dCacheRatio := decimal.NewFromFloat(cacheRatio)
@@ -412,22 +414,42 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 			dFileSearchQuota = decimal.NewFromFloat(fileSearchPrice).
 			dFileSearchQuota = decimal.NewFromFloat(fileSearchPrice).
 				Mul(decimal.NewFromInt(int64(fileSearchTool.CallCount))).
 				Mul(decimal.NewFromInt(int64(fileSearchTool.CallCount))).
 				Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
 				Div(decimal.NewFromInt(1000)).Mul(dGroupRatio).Mul(dQuotaPerUnit)
-			extraContent += fmt.Sprintf("File Search 调用 %d 次，调用花费 $%s",
+			extraContent += fmt.Sprintf("File Search 调用 %d 次，调用花费 %s",
 				fileSearchTool.CallCount, dFileSearchQuota.String())
 				fileSearchTool.CallCount, dFileSearchQuota.String())
 		}
 		}
 	}
 	}
 
 
 	var quotaCalculateDecimal decimal.Decimal
 	var quotaCalculateDecimal decimal.Decimal
+
+	var audioInputQuota decimal.Decimal
+	var audioInputPrice float64
 	if !priceData.UsePrice {
 	if !priceData.UsePrice {
-		nonCachedTokens := dPromptTokens.Sub(dCacheTokens)
-		cachedTokensWithRatio := dCacheTokens.Mul(dCacheRatio)
+		baseTokens := dPromptTokens
+		// 减去 cached tokens
+		var cachedTokensWithRatio decimal.Decimal
+		if !dCacheTokens.IsZero() {
+			baseTokens = baseTokens.Sub(dCacheTokens)
+			cachedTokensWithRatio = dCacheTokens.Mul(dCacheRatio)
+		}
+
+		// 减去 image tokens
+		var imageTokensWithRatio decimal.Decimal
+		if !dImageTokens.IsZero() {
+			baseTokens = baseTokens.Sub(dImageTokens)
+			imageTokensWithRatio = dImageTokens.Mul(dImageRatio)
+		}
 
 
-		promptQuota := nonCachedTokens.Add(cachedTokensWithRatio)
-		if imageTokens > 0 {
-			nonImageTokens := dPromptTokens.Sub(dImageTokens)
-			imageTokensWithRatio := dImageTokens.Mul(dImageRatio)
-			promptQuota = nonImageTokens.Add(imageTokensWithRatio)
+		// 减去 Gemini audio tokens
+		if !dAudioTokens.IsZero() {
+			audioInputPrice = operation_setting.GetGeminiInputAudioPricePerMillionTokens(modelName)
+			if audioInputPrice > 0 {
+				// 重新计算 base tokens
+				baseTokens = baseTokens.Sub(dAudioTokens)
+				audioInputQuota = decimal.NewFromFloat(audioInputPrice).Div(decimal.NewFromInt(1000000)).Mul(dAudioTokens).Mul(dGroupRatio).Mul(dQuotaPerUnit)
+				extraContent += fmt.Sprintf("Audio Input 花费 %s", audioInputQuota.String())
+			}
 		}
 		}
+		promptQuota := baseTokens.Add(cachedTokensWithRatio).Add(imageTokensWithRatio)
 
 
 		completionQuota := dCompletionTokens.Mul(dCompletionRatio)
 		completionQuota := dCompletionTokens.Mul(dCompletionRatio)
 
 
@@ -442,6 +464,8 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 	// 添加 responses tools call 调用的配额
 	// 添加 responses tools call 调用的配额
 	quotaCalculateDecimal = quotaCalculateDecimal.Add(dWebSearchQuota)
 	quotaCalculateDecimal = quotaCalculateDecimal.Add(dWebSearchQuota)
 	quotaCalculateDecimal = quotaCalculateDecimal.Add(dFileSearchQuota)
 	quotaCalculateDecimal = quotaCalculateDecimal.Add(dFileSearchQuota)
+	// 添加 audio input 独立计费
+	quotaCalculateDecimal = quotaCalculateDecimal.Add(audioInputQuota)
 
 
 	quota := int(quotaCalculateDecimal.Round(0).IntPart())
 	quota := int(quotaCalculateDecimal.Round(0).IntPart())
 	totalTokens := promptTokens + completionTokens
 	totalTokens := promptTokens + completionTokens
@@ -512,6 +536,11 @@ func postConsumeQuota(ctx *gin.Context, relayInfo *relaycommon.RelayInfo,
 			other["file_search_price"] = fileSearchPrice
 			other["file_search_price"] = fileSearchPrice
 		}
 		}
 	}
 	}
+	if !audioInputQuota.IsZero() {
+		other["audio_input_seperate_price"] = true
+		other["audio_input_token_count"] = audioTokens
+		other["audio_input_price"] = audioInputPrice
+	}
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, promptTokens, completionTokens, logModel,
 	model.RecordConsumeLog(ctx, relayInfo.UserId, relayInfo.ChannelId, promptTokens, completionTokens, logModel,
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
 		tokenName, quota, logContent, relayInfo.TokenId, userQuota, int(useTimeSeconds), relayInfo.IsStream, relayInfo.Group, other)
 }
 }

service/audio.go

@@ -3,6 +3,7 @@ package service
 import (
 import (
 	"encoding/base64"
 	"encoding/base64"
 	"fmt"
 	"fmt"
+	"strings"
 )
 )
 
 
 func parseAudio(audioBase64 string, format string) (duration float64, err error) {
 func parseAudio(audioBase64 string, format string) (duration float64, err error) {
@@ -29,3 +30,19 @@ func parseAudio(audioBase64 string, format string) (duration float64, err error)
 	duration = float64(samplesCount) / float64(sampleRate)
 	duration = float64(samplesCount) / float64(sampleRate)
 	return duration, nil
 	return duration, nil
 }
 }
+
+func DecodeBase64AudioData(audioBase64 string) (string, error) {
+	// 检查并移除 data:audio/xxx;base64, 前缀
+	idx := strings.Index(audioBase64, ",")
+	if idx != -1 {
+		audioBase64 = audioBase64[idx+1:]
+	}
+
+	// 解码 Base64 数据
+	_, err := base64.StdEncoding.DecodeString(audioBase64)
+	if err != nil {
+		return "", fmt.Errorf("base64 decode error: %v", err)
+	}
+
+	return audioBase64, nil
+}

setting/operation_setting/tools.go

@@ -14,6 +14,13 @@ const (
 	FileSearchPrice = 2.5
 	FileSearchPrice = 2.5
 )
 )
 
 
+const (
+	// Gemini Audio Input Price
+	Gemini25FlashPreviewInputAudioPrice     = 1.00
+	Gemini25FlashNativeAudioInputAudioPrice = 3.00
+	Gemini20FlashInputAudioPrice            = 0.70
+)
+
 func GetWebSearchPricePerThousand(modelName string, contextSize string) float64 {
 func GetWebSearchPricePerThousand(modelName string, contextSize string) float64 {
 	// 确定模型类型
 	// 确定模型类型
 	// https://platform.openai.com/docs/pricing Web search 价格按模型类型和 search context size 收费
 	// https://platform.openai.com/docs/pricing Web search 价格按模型类型和 search context size 收费
@@ -55,3 +62,14 @@ func GetWebSearchPricePerThousand(modelName string, contextSize string) float64
 func GetFileSearchPricePerThousand() float64 {
 func GetFileSearchPricePerThousand() float64 {
 	return FileSearchPrice
 	return FileSearchPrice
 }
 }
+
+func GetGeminiInputAudioPricePerMillionTokens(modelName string) float64 {
+	if strings.HasPrefix(modelName, "gemini-2.5-flash-preview") {
+		return Gemini25FlashPreviewInputAudioPrice
+	} else if strings.HasPrefix(modelName, "gemini-2.5-flash-preview-native-audio") {
+		return Gemini25FlashNativeAudioInputAudioPrice
+	} else if strings.HasPrefix(modelName, "gemini-2.0-flash") {
+		return Gemini20FlashInputAudioPrice
+	}
+	return 0
+}