Cyrene_v2/backend/services/voice/service.go

// Package voice provides speech-to-text and text-to-speech capabilities.
// It defines provider interfaces allowing pluggable backends (DashScope, Whisper, Edge-TTS, espeak).
package voice

import "context"

// STTProvider converts speech audio to text.
type STTProvider interface {
	// Name returns the provider identifier.
	Name() string // "dashscope", "whisper-local"

	// Recognize converts audio data to text.
	Recognize(ctx context.Context, audioData []byte) (string, error)
}

// TTSProvider converts text to speech audio.
type TTSProvider interface {
	// Name returns the provider identifier.
	Name() string // "edge-tts", "espeak", "silent"

	// Synthesize converts text to audio bytes.
	Synthesize(ctx context.Context, text string, voice string) ([]byte, error)
}

// Service manages STT and TTS providers with fallback chains.
type Service struct {
	sttProvider STTProvider
	ttsProviders []TTSProvider // Ordered by priority (first = preferred)
}

// Config holds voice service configuration.
type Config struct {
	STTProvider string // "dashscope" | "whisper-local"
	TTSProviders []string // ["edge-tts", "espeak", "silent"] ordered by priority
	DashScopeKey string
}

// New creates a new voice service.
func New(cfg Config) *Service {
	svc := &Service{}

	// STT: for now, use a simple placeholder
	if cfg.STTProvider == "dashscope" && cfg.DashScopeKey != "" {
		svc.sttProvider = &dashscopeSTT{apiKey: cfg.DashScopeKey}
	} else {
		svc.sttProvider = &silentSTT{}
	}

	// TTS: build fallback chain
	for _, name := range cfg.TTSProviders {
		switch name {
		case "edge-tts":
			svc.ttsProviders = append(svc.ttsProviders, &edgeTTS{})
		case "espeak":
			svc.ttsProviders = append(svc.ttsProviders, &espeakTTS{})
		case "silent":
			svc.ttsProviders = append(svc.ttsProviders, &silentTTS{})
		}
	}
	if len(svc.ttsProviders) == 0 {
		svc.ttsProviders = append(svc.ttsProviders, &silentTTS{})
	}

	return svc
}

// Recognize performs speech recognition with the configured STT provider.
func (s *Service) Recognize(ctx context.Context, audioData []byte) (string, error) {
	return s.sttProvider.Recognize(ctx, audioData)
}

// Synthesize performs TTS with automatic fallback through the provider chain.
func (s *Service) Synthesize(ctx context.Context, text string, voice string) ([]byte, error) {
	var lastErr error
	for _, p := range s.ttsProviders {
		audio, err := p.Synthesize(ctx, text, voice)
		if err == nil {
			return audio, nil
		}
		lastErr = err
	}
	return nil, lastErr
}

// --- Provider stubs ---

type dashscopeSTT struct{ apiKey string }
func (p *dashscopeSTT) Name() string { return "dashscope" }
func (p *dashscopeSTT) Recognize(_ context.Context, _ []byte) (string, error) {
	return "", nil // Placeholder: real implementation requires DashScope SDK
}

type silentSTT struct{}
func (p *silentSTT) Name() string { return "silent" }
func (p *silentSTT) Recognize(_ context.Context, _ []byte) (string, error) {
	return "", nil
}

type edgeTTS struct{}
func (p *edgeTTS) Name() string { return "edge-tts" }
func (p *edgeTTS) Synthesize(_ context.Context, text string, _ string) ([]byte, error) {
	return []byte(text), nil // Placeholder
}

type espeakTTS struct{}
func (p *espeakTTS) Name() string { return "espeak" }
func (p *espeakTTS) Synthesize(_ context.Context, text string, _ string) ([]byte, error) {
	return []byte(text), nil
}

type silentTTS struct{}
func (p *silentTTS) Name() string { return "silent" }
func (p *silentTTS) Synthesize(_ context.Context, _ string, _ string) ([]byte, error) {
	return make([]byte, 0), nil
}