CamTalk/backend/internal/ai/tts/openai.go

package tts

import (
	"bytes"
	"context"
	"encoding/json"
	"fmt"
	"io"
	"net/http"
	"time"

	"go.uber.org/zap"
)

// OpenAIService 基于 OpenAI TTS API 的语音合成实现。
type OpenAIService struct {
	apiKey   string
	voice    string
	speed    float64
	endpoint string
	timeout  time.Duration
	logger   *zap.SugaredLogger
	client   *http.Client
}

// NewOpenAIService 创建 OpenAI TTS 服务。
func NewOpenAIService(apiKey, voice, endpoint string, speed float64, timeoutSec int, logger *zap.SugaredLogger) *OpenAIService {
	if voice == "" {
		voice = "alloy"
	}
	if speed <= 0 {
		speed = 1.0
	}
	if endpoint == "" {
		endpoint = "https://api.openai.com/v1"
	}
	timeout := time.Duration(timeoutSec) * time.Second
	if timeout <= 0 {
		timeout = 5 * time.Second
	}
	return &OpenAIService{
		apiKey:   apiKey,
		voice:    voice,
		speed:    speed,
		endpoint: endpoint,
		timeout:  timeout,
		logger:   logger,
		client:   &http.Client{Timeout: 30 * time.Second},
	}
}

// ttsRequest OpenAI TTS API 请求。
type ttsRequest struct {
	Model          string  `json:"model"`
	Input          string  `json:"input"`
	Voice          string  `json:"voice"`
	ResponseFormat string  `json:"response_format"`
	Speed          float64 `json:"speed"`
}

// SynthesizeStream 实现 tts.Service。从 textStream 读取句子，逐句调用 OpenAI TTS API。
func (o *OpenAIService) SynthesizeStream(ctx context.Context, textStream <-chan string, opts Options) (<-chan Chunk, error) {
	voice := opts.Voice
	if voice == "" {
		voice = o.voice
	}
	speed := opts.Speed
	if speed <= 0 {
		speed = o.speed
	}

	ch := make(chan Chunk, 4)
	go func() {
		defer close(ch)

		for text := range textStream {
			if text == "" {
				continue
			}

			audio, err := o.synthesize(ctx, text, voice, speed)
			if err != nil {
				o.logger.Warnw("tts: synthesize failed", "error", err, "text", text)
				// 静默跳过，不中断整个流
				continue
			}

			select {
			case ch <- Chunk{Audio: audio, IsLast: false}:
			case <-ctx.Done():
				return
			}
		}

		// textStream 关闭，发送 IsLast 标记
		select {
		case ch <- Chunk{Audio: nil, IsLast: true}:
		case <-ctx.Done():
		}
	}()

	return ch, nil
}

// synthesize 调用 OpenAI TTS API 合成单个句子。
func (o *OpenAIService) synthesize(ctx context.Context, text, voice string, speed float64) ([]byte, error) {
	// 单句超时
	ctx, cancel := context.WithTimeout(ctx, o.timeout)
	defer cancel()

	body := ttsRequest{
		Model:          "tts-1",
		Input:          text,
		Voice:          voice,
		ResponseFormat: "mp3",
		Speed:          speed,
	}

	payload, err := json.Marshal(body)
	if err != nil {
		return nil, fmt.Errorf("tts: marshal request: %w", err)
	}

	req, err := http.NewRequestWithContext(ctx, http.MethodPost, o.endpoint+"/audio/speech", bytes.NewReader(payload))
	if err != nil {
		return nil, fmt.Errorf("tts: create request: %w", err)
	}
	req.Header.Set("Content-Type", "application/json")
	req.Header.Set("Authorization", "Bearer "+o.apiKey)

	resp, err := o.client.Do(req)
	if err != nil {
		return nil, fmt.Errorf("tts: send request: %w", err)
	}
	defer resp.Body.Close()

	if resp.StatusCode != http.StatusOK {
		errBody, _ := io.ReadAll(resp.Body)
		return nil, fmt.Errorf("tts: api error (status %d): %s", resp.StatusCode, string(errBody))
	}

	// 读取整个 MP3 响应
	audio, err := io.ReadAll(resp.Body)
	if err != nil {
		return nil, fmt.Errorf("tts: read response: %w", err)
	}

	return audio, nil
}