迁移音频生成与语音合成到 gateway 并补充 simulation 测试

2026-06-07 10:26:57 +08:00
parent 78ab867a9f
commit dc14866210
22 changed files with 2475 additions and 55 deletions
@@ -2,8 +2,11 @@ package clients

 import (
 	"context"
+	"encoding/base64"
+	"encoding/hex"
 	"net/http"
 	"strings"
+	"time"
 )

 type JimengClient struct{ HTTPClient *http.Client }
@@ -15,6 +18,7 @@ type MidjourneyClient struct{ HTTPClient *http.Client }
 type ViduClient struct{ HTTPClient *http.Client }
 type AliyunBailianClient struct{ HTTPClient *http.Client }
 type NewAPIClient struct{ HTTPClient *http.Client }
+type SunoClient struct{ HTTPClient *http.Client }

 func (c JimengClient) Run(ctx context.Context, request Request) (Response, error) {
 	return providerTaskClient{HTTPClient: c.HTTPClient, Spec: jimengSpec()}.Run(ctx, request)
@@ -33,6 +37,9 @@ func (c HunyuanVideoClient) Run(ctx context.Context, request Request) (Response,
 }

 func (c MinimaxClient) Run(ctx context.Context, request Request) (Response, error) {
+	if request.Kind == "speech.generations" {
+		return c.runSpeech(ctx, request)
+	}
 	return providerTaskClient{HTTPClient: c.HTTPClient, Spec: minimaxSpec()}.Run(ctx, request)
 }

@@ -52,6 +59,10 @@ func (c NewAPIClient) Run(ctx context.Context, request Request) (Response, error
 	return providerTaskClient{HTTPClient: c.HTTPClient, Spec: newAPISpec()}.Run(ctx, request)
 }

+func (c SunoClient) Run(ctx context.Context, request Request) (Response, error) {
+	return providerTaskClient{HTTPClient: c.HTTPClient, Spec: sunoSpec()}.Run(ctx, request)
+}
+
 func jimengSpec() providerTaskSpec {
 	return providerTaskSpec{
 		Name: "jimeng",
@@ -149,6 +160,114 @@ func minimaxSpec() providerTaskSpec {
 	}
 }

+func (c MinimaxClient) runSpeech(ctx context.Context, request Request) (Response, error) {
+	startedAt := time.Now()
+	payload := minimaxSpeechPayload(request)
+	result, requestID, err := providerPostJSON(ctx, httpClient(request.HTTPClient, c.HTTPClient), providerURL(request.Candidate.BaseURL, "/t2a_v2"), payload, request.Candidate.Credentials, "bearer")
+	finishedAt := time.Now()
+	if err != nil {
+		return Response{}, annotateResponseError(err, requestID, startedAt, finishedAt)
+	}
+	audioHex := strings.TrimSpace(stringFromPathValue(valueAtPath(result, "data.audio")))
+	if audioHex == "" {
+		message := firstNonEmptyString(valueAtPath(result, "base_resp.status_msg"), valueAtPath(result, "message"), "minimax speech audio is missing")
+		return Response{}, &ClientError{Code: "invalid_response", Message: message, RequestID: firstNonEmptyString(requestID, requestIDFromResult(result)), ResponseStartedAt: startedAt, ResponseFinishedAt: finishedAt, ResponseDurationMS: responseDurationMS(startedAt, finishedAt), Retryable: false}
+	}
+	audioBytes, err := hex.DecodeString(audioHex)
+	if err != nil {
+		return Response{}, &ClientError{Code: "invalid_response", Message: "minimax speech audio hex is invalid: " + err.Error(), RequestID: firstNonEmptyString(requestID, requestIDFromResult(result)), ResponseStartedAt: startedAt, ResponseFinishedAt: finishedAt, ResponseDurationMS: responseDurationMS(startedAt, finishedAt), Retryable: false}
+	}
+	normalized := cloneMapAny(result)
+	normalized["status"] = "success"
+	normalized["created"] = time.Now().UnixMilli()
+	normalized["model"] = request.Model
+	normalized["raw_data"] = cloneMapAny(result)
+	normalized["data"] = []any{map[string]any{
+		"type":      "audio",
+		"content":   "data:audio/mpeg;base64," + base64.StdEncoding.EncodeToString(audioBytes),
+		"mime_type": "audio/mpeg",
+		"uploaded":  false,
+	}}
+	return Response{
+		Result:             normalized,
+		RequestID:          firstNonEmptyString(requestID, requestIDFromResult(result)),
+		Progress:           providerProgress(request),
+		ResponseStartedAt:  startedAt,
+		ResponseFinishedAt: finishedAt,
+		ResponseDurationMS: responseDurationMS(startedAt, finishedAt),
+	}, nil
+}
+
+func minimaxSpeechPayload(request Request) map[string]any {
+	body := cloneBody(request.Body)
+	body["model"] = upstreamModelName(request.Candidate)
+	voiceID := firstNonEmptyString(body["voice_id"], body["voiceId"])
+	speed := firstPresent(body["speed"], float64(1))
+	vol := firstPresent(body["vol"], body["volume"], float64(1))
+	pitch := firstPresent(body["pitch"], float64(0))
+	voiceSetting := map[string]any{
+		"voice_id": voiceID,
+		"speed":    speed,
+		"vol":      vol,
+		"pitch":    pitch,
+	}
+	if emotion := firstNonEmptyString(body["emotion"]); emotion != "" {
+		voiceSetting["emotion"] = emotion
+	}
+	delete(body, "voice_id")
+	delete(body, "voiceId")
+	delete(body, "speed")
+	delete(body, "vol")
+	delete(body, "volume")
+	delete(body, "pitch")
+	delete(body, "emotion")
+	body["voice_setting"] = voiceSetting
+	return body
+}
+
+func sunoSpec() providerTaskSpec {
+	return providerTaskSpec{
+		Name:       "suno",
+		SubmitPath: func(Request, map[string]any) string { return "/generator/suno" },
+		PollPath: func(_ Request, upstreamTaskID string, _ map[string]any) string {
+			return "/v2/sunoinfo?id=" + upstreamTaskID
+		},
+		Auth:            "bearer",
+		TaskIDPaths:     []string{"data"},
+		StatusPaths:     []string{"data.status"},
+		SuccessStatuses: []string{"succeeded", "complete", "completed"},
+		FailureStatuses: []string{"failed"},
+		DefaultSubmitBody: func(request Request, body map[string]any) map[string]any {
+			body["task"] = "create"
+			body["model"] = sunoMappedModel(upstreamModelName(request.Candidate))
+			if body["customMode"] == nil {
+				body["customMode"] = false
+			}
+			if body["makeInstrumental"] == nil {
+				body["makeInstrumental"] = false
+			}
+			return body
+		},
+	}
+}
+
+func sunoMappedModel(model string) string {
+	switch strings.TrimSpace(model) {
+	case "chirp-v3-0", "chirp-v3-5":
+		return "v40"
+	case "chirp-v4-0":
+		return "v40"
+	case "chirp-v4-5":
+		return "v45"
+	case "chirp-v4-5+":
+		return "v45+"
+	case "chirp-v5-0":
+		return "v50"
+	default:
+		return model
+	}
+}
+
 func midjourneySpec() providerTaskSpec {
 	return providerTaskSpec{
 		Name: "midjourney",