package controller import ( "bytes" "context" "encoding/json" "errors" "fmt" "github.com/gin-gonic/gin" "io" "net/http" "one-api/common" "one-api/model" "strings" ) var availableVoices = []string{ "alloy", "echo", "fable", "onyx", "nova", "shimmer", } func relayAudioHelper(c *gin.Context, relayMode int) *OpenAIErrorWithStatusCode { tokenId := c.GetInt("token_id") channelType := c.GetInt("channel") channelId := c.GetInt("channel_id") userId := c.GetInt("id") group := c.GetString("group") var audioRequest AudioRequest if !strings.HasPrefix(c.Request.URL.Path, "/v1/audio/transcriptions") { err := common.UnmarshalBodyReusable(c, &audioRequest) if err != nil { return errorWrapper(err, "bind_request_body_failed", http.StatusBadRequest) } } else { audioRequest = AudioRequest{ Model: "whisper-1", } } //err := common.UnmarshalBodyReusable(c, &audioRequest) // request validation if audioRequest.Model == "" { return errorWrapper(errors.New("model is required"), "required_field_missing", http.StatusBadRequest) } if strings.HasPrefix(audioRequest.Model, "tts-1") { if audioRequest.Voice == "" { return errorWrapper(errors.New("voice is required"), "required_field_missing", http.StatusBadRequest) } if !common.StringsContains(availableVoices, audioRequest.Voice) { return errorWrapper(errors.New("voice must be one of "+strings.Join(availableVoices, ", ")), "invalid_field_value", http.StatusBadRequest) } } preConsumedTokens := common.PreConsumedQuota modelRatio := common.GetModelRatio(audioRequest.Model) groupRatio := common.GetGroupRatio(group) ratio := modelRatio * groupRatio preConsumedQuota := int(float64(preConsumedTokens) * ratio) userQuota, err := model.CacheGetUserQuota(userId) if err != nil { return errorWrapper(err, "get_user_quota_failed", http.StatusInternalServerError) } if userQuota-preConsumedQuota < 0 { return errorWrapper(errors.New("user quota is not enough"), "insufficient_user_quota", http.StatusForbidden) } err = model.CacheDecreaseUserQuota(userId, preConsumedQuota) if err != nil { return errorWrapper(err, "decrease_user_quota_failed", http.StatusInternalServerError) } if userQuota > 100*preConsumedQuota { // in this case, we do not pre-consume quota // because the user has enough quota preConsumedQuota = 0 } if preConsumedQuota > 0 { userQuota, err = model.PreConsumeTokenQuota(tokenId, preConsumedQuota) if err != nil { return errorWrapper(err, "pre_consume_token_quota_failed", http.StatusForbidden) } } // map model name modelMapping := c.GetString("model_mapping") if modelMapping != "" { modelMap := make(map[string]string) err := json.Unmarshal([]byte(modelMapping), &modelMap) if err != nil { return errorWrapper(err, "unmarshal_model_mapping_failed", http.StatusInternalServerError) } if modelMap[audioRequest.Model] != "" { audioRequest.Model = modelMap[audioRequest.Model] } } baseURL := common.ChannelBaseURLs[channelType] requestURL := c.Request.URL.String() if c.GetString("base_url") != "" { baseURL = c.GetString("base_url") } fullRequestURL := getFullRequestURL(baseURL, requestURL, channelType) requestBody := c.Request.Body req, err := http.NewRequest(c.Request.Method, fullRequestURL, requestBody) if err != nil { return errorWrapper(err, "new_request_failed", http.StatusInternalServerError) } req.Header.Set("Authorization", c.Request.Header.Get("Authorization")) req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type")) req.Header.Set("Accept", c.Request.Header.Get("Accept")) resp, err := httpClient.Do(req) if err != nil { return errorWrapper(err, "do_request_failed", http.StatusInternalServerError) } err = req.Body.Close() if err != nil { return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError) } err = c.Request.Body.Close() if err != nil { return errorWrapper(err, "close_request_body_failed", http.StatusInternalServerError) } if resp.StatusCode != http.StatusOK { return relayErrorHandler(resp) } var audioResponse AudioResponse defer func(ctx context.Context) { go func() { quota := 0 var promptTokens = 0 if strings.HasPrefix(audioRequest.Model, "tts-1") { quota = countAudioToken(audioRequest.Input, audioRequest.Model) promptTokens = quota } else { quota = countAudioToken(audioResponse.Text, audioRequest.Model) } quota = int(float64(quota) * ratio) if ratio != 0 && quota <= 0 { quota = 1 } quotaDelta := quota - preConsumedQuota err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota, true) if err != nil { common.SysError("error consuming token remain quota: " + err.Error()) } err = model.CacheUpdateUserQuota(userId) if err != nil { common.SysError("error update user quota cache: " + err.Error()) } if quota != 0 { tokenName := c.GetString("token_name") logContent := fmt.Sprintf("模型倍率 %.2f,分组倍率 %.2f", modelRatio, groupRatio) model.RecordConsumeLog(ctx, userId, channelId, promptTokens, 0, audioRequest.Model, tokenName, quota, logContent, tokenId, userQuota) model.UpdateUserUsedQuotaAndRequestCount(userId, quota) channelId := c.GetInt("channel_id") model.UpdateChannelUsedQuota(channelId, quota) } }() }(c.Request.Context()) responseBody, err := io.ReadAll(resp.Body) if err != nil { return errorWrapper(err, "read_response_body_failed", http.StatusInternalServerError) } err = resp.Body.Close() if err != nil { return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError) } if strings.HasPrefix(audioRequest.Model, "tts-1") { } else { err = json.Unmarshal(responseBody, &audioResponse) if err != nil { return errorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError) } } resp.Body = io.NopCloser(bytes.NewBuffer(responseBody)) for k, v := range resp.Header { c.Writer.Header().Set(k, v[0]) } c.Writer.WriteHeader(resp.StatusCode) _, err = io.Copy(c.Writer, resp.Body) if err != nil { return errorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError) } err = resp.Body.Close() if err != nil { return errorWrapper(err, "close_response_body_failed", http.StatusInternalServerError) } return nil }