package microsoft import ( "bytes" "context" "encoding/json" "errors" "fmt" "io" "log" "net/http" "strings" "sync" "time" "tts/internal/config" "tts/internal/models" "tts/internal/utils" ) const ( userAgent = "okhttp/4.5.0" voicesEndpoint = "https://%s.tts.speech.microsoft.com/cognitiveservices/voices/list" ttsEndpoint = "https://%s.tts.speech.microsoft.com/cognitiveservices/v1" ssmlTemplate = ` %s ` ) // Client 是Microsoft TTS API的客户端实现 type Client struct { defaultVoice string defaultRate string defaultPitch string defaultFormat string maxTextLength int httpClient *http.Client voicesCache []models.Voice voicesCacheMu sync.RWMutex voicesCacheExpiry time.Time // 端点和认证信息 endpoint map[string]interface{} endpointMu sync.RWMutex endpointExpiry time.Time ssmProcessor *config.SSMLProcessor } // NewClient 创建一个新的Microsoft TTS客户端 func NewClient(cfg *config.Config) *Client { // 从Viper配置中创建SSML处理器 ssmProcessor, err := config.NewSSMLProcessor(&cfg.SSML) if err != nil { log.Fatalf("创建SSML处理器失败: %v", err) } client := &Client{ defaultVoice: cfg.TTS.DefaultVoice, defaultRate: cfg.TTS.DefaultRate, defaultPitch: cfg.TTS.DefaultPitch, defaultFormat: cfg.TTS.DefaultFormat, maxTextLength: cfg.TTS.MaxTextLength, httpClient: &http.Client{ Timeout: time.Duration(cfg.TTS.RequestTimeout) * time.Second, }, voicesCacheExpiry: time.Time{}, // 初始时缓存为空 endpointExpiry: time.Time{}, // 初始时端点为空 ssmProcessor: ssmProcessor, } return client } // getEndpoint 获取或刷新认证端点 func (c *Client) getEndpoint(ctx context.Context) (map[string]interface{}, error) { c.endpointMu.RLock() if !c.endpointExpiry.IsZero() && time.Now().Before(c.endpointExpiry) && c.endpoint != nil { endpoint := c.endpoint c.endpointMu.RUnlock() return endpoint, nil } c.endpointMu.RUnlock() // 获取新的端点信息 endpoint, err := utils.GetEndpoint() if err != nil { return nil, err } // 更新缓存 c.endpointMu.Lock() c.endpoint = endpoint c.endpointExpiry = time.Now().Add(30 * time.Minute) // 令牌有效期通常是1小时，提前刷新 c.endpointMu.Unlock() return endpoint, nil } // ListVoices 获取可用的语音列表 func (c *Client) ListVoices(ctx context.Context, locale string) ([]models.Voice, error) { // 检查缓存是否有效 c.voicesCacheMu.RLock() if !c.voicesCacheExpiry.IsZero() && time.Now().Before(c.voicesCacheExpiry) && len(c.voicesCache) > 0 { voices := c.voicesCache c.voicesCacheMu.RUnlock() // 如果指定了locale，则过滤结果 if locale != "" { var filtered []models.Voice for _, voice := range voices { if strings.HasPrefix(voice.Locale, locale) { filtered = append(filtered, voice) } } return filtered, nil } return voices, nil } c.voicesCacheMu.RUnlock() // 缓存无效，需要从API获取 endpoint, err := c.getEndpoint(ctx) if err != nil { return nil, err } url := fmt.Sprintf(voicesEndpoint, endpoint["r"]) req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil) if err != nil { return nil, err } // 使用新的认证方式 req.Header.Set("Authorization", endpoint["t"].(string)) resp, err := c.httpClient.Do(req) if err != nil { return nil, err } defer resp.Body.Close() if resp.StatusCode != http.StatusOK { body, _ := io.ReadAll(resp.Body) return nil, fmt.Errorf("API error: %s, status: %d", string(body), resp.StatusCode) } var msVoices []MicrosoftVoice if err := json.NewDecoder(resp.Body).Decode(&msVoices); err != nil { return nil, err } // 转换为通用模型 voices := make([]models.Voice, len(msVoices)) for i, v := range msVoices { voices[i] = models.Voice{ Name: v.Name, DisplayName: v.DisplayName, LocalName: v.LocalName, ShortName: v.ShortName, Gender: v.Gender, Locale: v.Locale, LocaleName: v.LocaleName, StyleList: v.StyleList, SampleRateHertz: v.SampleRateHertz, // 直接使用字符串，无需转换 } } // 更新缓存 c.voicesCacheMu.Lock() c.voicesCache = voices c.voicesCacheExpiry = time.Now().Add(1 * time.Hour) // 缓存1小时 c.voicesCacheMu.Unlock() // 如果指定了locale，则过滤结果 if locale != "" { var filtered []models.Voice for _, voice := range voices { if strings.HasPrefix(voice.Locale, locale) { filtered = append(filtered, voice) } } return filtered, nil } return voices, nil } // SynthesizeSpeech 将文本转换为语音 func (c *Client) SynthesizeSpeech(ctx context.Context, req models.TTSRequest) (*models.TTSResponse, error) { resp, err := c.createTTSRequest(ctx, req) if err != nil { return nil, err } defer resp.Body.Close() // 读取音频数据 audio, err := io.ReadAll(resp.Body) if err != nil { return nil, err } return &models.TTSResponse{ AudioContent: audio, ContentType: "audio/mpeg", CacheHit: false, }, nil } // createTTSRequest 创建并执行TTS请求，返回HTTP响应 func (c *Client) createTTSRequest(ctx context.Context, req models.TTSRequest) (*http.Response, error) { // 参数验证 if req.Text == "" { return nil, errors.New("文本不能为空") } if len(req.Text) > c.maxTextLength { return nil, fmt.Errorf("文本长度超过限制 (%d > %d)", len(req.Text), c.maxTextLength) } // 使用默认值填充空白参数 voice := req.Voice if voice == "" { voice = c.defaultVoice } style := req.Style if req.Style == "" { style = "general" } rate := req.Rate if rate == "" { rate = c.defaultRate } pitch := req.Pitch if pitch == "" { pitch = c.defaultPitch } // 提取语言 locale := "zh-CN" // 默认 parts := strings.Split(voice, "-") if len(parts) >= 2 { locale = parts[0] + "-" + parts[1] } // 对文本进行HTML转义，防止XML解析错误 escapedText := c.ssmProcessor.EscapeSSML(req.Text) // 准备SSML内容 ssml := fmt.Sprintf(ssmlTemplate, locale, voice, style, rate, pitch, escapedText) // 获取端点信息 endpoint, err := c.getEndpoint(ctx) if err != nil { return nil, err } // 准备请求 url := fmt.Sprintf(ttsEndpoint, endpoint["r"]) reqBody := bytes.NewBufferString(ssml) httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, reqBody) if err != nil { return nil, err } httpReq.Header.Set("Authorization", endpoint["t"].(string)) httpReq.Header.Set("Content-Type", "application/ssml+xml") httpReq.Header.Set("X-Microsoft-OutputFormat", c.defaultFormat) httpReq.Header.Set("User-Agent", userAgent) // 发送请求 resp, err := c.httpClient.Do(httpReq) if err != nil { return nil, err } if resp.StatusCode != http.StatusOK { // 获取响应体以便调试 body, _ := io.ReadAll(resp.Body) resp.Body.Close() log.Printf("TTS API错误: %s, 状态码: %d", string(body), resp.StatusCode) return nil, fmt.Errorf("TTS API错误: %s, 状态码: %d", string(body), resp.StatusCode) } return resp, nil }