Cyrene/backend/voice-service/internal/service/stt_service.go

package service

import (
	"context"
	"fmt"
	"os"
	"os/exec"
	"path/filepath"
	"strings"
	"time"

	"github.com/yourname/cyrene-ai/voice-service/internal/config"
)

// SupportedLanguages STT 支持的语言列表
var SupportedLanguages = []string{"zh", "en", "ja", "ko", "auto"}

// STTService 语音转文字服务。
// 优先使用 DashScope API，不可用时回退到本地 Whisper。
type STTService struct {
	whisperBinary string
	whisperModel  string
	language      string
	dashscope     *DashScopeSTT // 实时 ASR (qwen3-asr-flash-realtime)
}

// NewSTTService 创建 STT 服务。
func NewSTTService(cfg *config.Config) *STTService {
	// 实时模型用于所有 WebSocket ASR 请求（支持 one-shot 和 streaming）
	// 离线模型 (qwen3-asr-flash-2026-02-10) 是 HTTP REST API，暂未实现
	model := cfg.DashScopeSTTRealtime
	if model == "" {
		model = cfg.DashScopeModel
	}
	return &STTService{
		whisperBinary: cfg.WhisperBinary,
		whisperModel:  cfg.WhisperModel,
		language:      cfg.WhisperLanguage,
		dashscope:     NewDashScopeSTT(cfg.DashScopeAPIKey, model),
	}
}

// IsAvailable 检查是否有任一 STT 引擎可用。
func (s *STTService) IsAvailable() bool {
	if s.dashscope.IsAvailable() {
		return true
	}
	_, err := os.Stat(s.whisperBinary)
	return err == nil
}

// Transcribe 将音频数据转录为文字。
// 优先使用 DashScope，不可用时回退到本地 Whisper。
func (s *STTService) Transcribe(audioData []byte, format string, language string) (string, error) {
	if language == "" {
		language = s.language
	}
	if !isSupportedLanguage(language) {
		return "", fmt.Errorf("不支持的语言: %s，支持的语言: %s", language, strings.Join(SupportedLanguages, ", "))
	}

	// 优先 DashScope
	if s.dashscope.IsAvailable() {
		ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
		defer cancel()
		text, err := s.dashscope.Transcribe(ctx, audioData, format, language)
		if err == nil {
			return text, nil
		}
		// DashScope 失败，返回具体错误而不是回退到 Whisper
		return "", fmt.Errorf("语音识别失败: %w", err)
	}

	// 回退到本地 Whisper
	return s.transcribeWhisper(audioData, format, language)
}

// StartStreaming 创建持久的流式语音识别会话。
func (s *STTService) StartStreaming(format, language string) (*StreamingSession, error) {
	if !s.dashscope.IsAvailable() {
		return nil, fmt.Errorf("流式识别需要 DashScope，请配置 DASHSCOPE_API_KEY")
	}
	if language == "" {
		language = s.language
	}
	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
	defer cancel()
	return s.dashscope.StartStreaming(ctx, format, language)
}

// transcribeWhisper 使用本地 Whisper 引擎转录。
func (s *STTService) transcribeWhisper(audioData []byte, format string, language string) (string, error) {
	if _, err := os.Stat(s.whisperBinary); err != nil {
		return "", fmt.Errorf("STT 引擎不可用: DashScope API Key 未配置且 Whisper 未安装")
	}

	ext := normalizeExt(format)
	tmpFile, err := os.CreateTemp("/tmp", "cyrene-stt-*"+ext)
	if err != nil {
		return "", fmt.Errorf("创建临时文件失败: %w", err)
	}
	tmpPath := tmpFile.Name()
	defer os.Remove(tmpPath)

	if _, err := tmpFile.Write(audioData); err != nil {
		tmpFile.Close()
		return "", fmt.Errorf("写入临时文件失败: %w", err)
	}
	tmpFile.Close()

	inputPath := tmpPath
	if format != "wav" && format != "" {
		convertedPath := tmpPath + ".wav"
		if err := convertToWav(tmpPath, convertedPath); err == nil {
			defer os.Remove(convertedPath)
			inputPath = convertedPath
		}
	}

	outputPrefix := strings.TrimSuffix(inputPath, filepath.Ext(inputPath))
	outputTxt := outputPrefix + ".txt"

	cmd := exec.Command(s.whisperBinary,
		"-m", s.whisperModel,
		"-l", language,
		"-f", inputPath,
		"-otxt",
		"-of", outputPrefix,
	)
	cmd.Stderr = os.Stderr

	if err := cmd.Run(); err != nil {
		os.Remove(outputTxt)
		return "", fmt.Errorf("whisper 转录失败: %w", err)
	}

	defer os.Remove(outputTxt)
	txtData, err := os.ReadFile(outputTxt)
	if err != nil {
		return "", fmt.Errorf("读取转录结果失败: %w", err)
	}
	return strings.TrimSpace(string(txtData)), nil
}

// GetStatus 返回服务状态。
func (s *STTService) GetStatus() map[string]interface{} {
	binaryAvailable := false
	if _, err := os.Stat(s.whisperBinary); err == nil {
		binaryAvailable = true
	}
	modelExists := false
	if _, err := os.Stat(s.whisperModel); err == nil {
		modelExists = true
	}

	return map[string]interface{}{
		"available":  s.IsAvailable(),
		"primary":    "dashscope",
		"dashscope":  s.dashscope.GetStatus(),
		"whisper": map[string]interface{}{
			"available":        binaryAvailable && modelExists,
			"binary_available": binaryAvailable,
			"model_loaded":     modelExists,
			"model_name":       filepath.Base(s.whisperModel),
		},
		"default_language":    s.language,
		"supported_languages": SupportedLanguages,
	}
}

// normalizeExt 规范化文件扩展名。
func normalizeExt(format string) string {
	switch strings.ToLower(format) {
	case "wav":
		return ".wav"
	case "mp3", "mpeg":
		return ".mp3"
	case "ogg", "opus":
		return ".ogg"
	case "flac":
		return ".flac"
	case "m4a", "mp4", "aac":
		return ".m4a"
	default:
		return ".wav"
	}
}

// isSupportedLanguage 检查语言是否支持。
func isSupportedLanguage(lang string) bool {
	for _, l := range SupportedLanguages {
		if l == lang {
			return true
		}
	}
	return false
}

// convertToWav 使用 ffmpeg 将音频转换为 WAV 格式。
func convertToWav(inputPath, outputPath string) error {
	cmd := exec.Command("ffmpeg",
		"-i", inputPath,
		"-ar", "16000",
		"-ac", "1",
		"-c:a", "pcm_s16le",
		outputPath,
		"-y",
	)
	cmd.Stderr = nil
	return cmd.Run()
}