feat(voice): real TTS fallback chain and whisper-compatible STT

- TTS providers: cloud (OpenAI-compatible /audio/speech, e.g. Bailian
  compatible mode with cosyvoice), local self-hosted HTTP server,
  edge-tts CLI, espeak CLI, silent — ordered by TTS_PROVIDER list,
  skipping providers whose binary/config is missing
- STT: whisper-compatible /audio/transcriptions endpoint (cloud or
  self-hosted faster-whisper), replacing the empty dashscope stub
- config: TTS_CLOUD_URL/KEY/MODEL/VOICE, TTS_LOCAL_URL, STT_API_URL/
  KEY/MODEL; cloud keys fall back to DASHSCOPE_API_KEY
- gateway /voice/tts now returns real audio/mpeg bytes instead of a
  byte count
- tests: cloud/local/whisper round-trips via httptest, chain skip
  behavior, fallback order; kept and adapted prior service tests

backend/cmd/cyrene/main.go

@@ -10,6 +10,7 @@ import (
 	"net/http"
 	"os"
 	"os/signal"
+	"strings"
 	"syscall"
 	"time"
 
@@ -124,12 +125,24 @@ func main() {
 	logger.Infof("[plugin] loaded %d plugins: %v", reg.Count(), reg.ListNames())
 
 	// === 5. Voice Service ===
+	orEmpty := func(v, fallback string) string {
+		if v != "" {
+			return v
+		}
+		return fallback
+	}
 	voiceSvc := voicesvc.New(voicesvc.Config{
-		STTProvider:  cfg.ASRProvider,
-		TTSProviders: []string{cfg.TTSProvider},
-		DashScopeKey: cfg.DashScopeAPIKey,
+		STTApiURL:     cfg.STTApiURL,
+		STTApiKey:     orEmpty(cfg.STTApiKey, cfg.DashScopeAPIKey),
+		STTModel:      cfg.STTModel,
+		TTSProviders:  splitComma(cfg.TTSProvider),
+		TTSCloudURL:   cfg.TTSCloudURL,
+		TTSCloudKey:   orEmpty(cfg.TTSCloudKey, cfg.DashScopeAPIKey),
+		TTSCloudModel: cfg.TTSCloudModel,
+		TTSCloudVoice: cfg.TTSCloudVoice,
+		TTSLocalURL:   cfg.TTSLocalURL,
 	})
-	logger.Infof("[voice] ready")
+	logger.Infof("[voice] ready (tts chain: %v)", voiceSvc.TTSProviderNames())
 
 	// === 6. Memory Service ===
 	var memSvc *memory.Service
@@ -407,6 +420,17 @@ type llmLoggerAdapter struct{ *llmlog.CallLogger }
 
 func (a *llmLoggerAdapter) Recent(n int) []map[string]any { return a.CallLogger.RecentMaps(n) }
 
+// splitComma splits a comma-separated config value into trimmed non-empty parts.
+func splitComma(s string) []string {
+	var out []string
+	for _, p := range strings.Split(s, ",") {
+		if p = strings.TrimSpace(p); p != "" {
+			out = append(out, p)
+		}
+	}
+	return out
+}
+
 func pluginNamesToMap(names []string) []map[string]string {
 	result := make([]map[string]string, len(names))
 	for i, n := range names {