feat: 语音流式输入管线 + VAD前端集成 + 插件-工具合并清理

- 前端: VAD语音检测(@ricky0123/vad-web) + useVoiceInput双模式(流式WS/REST)
- Gateway: VoiceStreamManager代理WS流式STT到voice-service
- Voice-service: DashScope REST → Realtime WS → Whisper三级引擎 + ffmpeg转码
- 共享模块: pkg/audio(音频转换) + pkg/dashscope(ASR REST客户端)
- 清理: 移除旧plugin-manager和pkg/plugins，完成插件→工具合并
- 文档: 完善gateway-api.md和voice-service.md语音API文档
- 工具: scripts/voice/ 语音转换脚本集

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/ai-core/Dockerfile

@@ -19,7 +19,7 @@ RUN CGO_ENABLED=0 GOOS=linux go build -ldflags="-s -w" -o /ai-core ./cmd/main.go
 # ========== 运行阶段 ==========
 FROM alpine:3.20
 
-RUN apk add --no-cache ca-certificates tzdata && \
+RUN apk add --no-cache ca-certificates tzdata ffmpeg && \
     cp /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && \
     echo "Asia/Shanghai" > /etc/timezone
 

backend/ai-core/cmd/main.go

@@ -29,19 +29,19 @@ import (
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/subsession"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/tools"
 
-	plgManager "git.yeij.top/AskaEth/Cyrene/pkg/plugins/manager"
-	plgSDK "git.yeij.top/AskaEth/Cyrene/pkg/plugins/sdk"
-	pluginCalc "git.yeij.top/AskaEth/Cyrene/pkg/plugins/calculator"
-	pluginCrypto "git.yeij.top/AskaEth/Cyrene/pkg/plugins/crypto"
-	pluginDate "git.yeij.top/AskaEth/Cyrene/pkg/plugins/datetime"
-	pluginFile "git.yeij.top/AskaEth/Cyrene/pkg/plugins/file"
-	pluginHTTP "git.yeij.top/AskaEth/Cyrene/pkg/plugins/http"
-	pluginJSON "git.yeij.top/AskaEth/Cyrene/pkg/plugins/json"
-	pluginMD "git.yeij.top/AskaEth/Cyrene/pkg/plugins/markdown"
-	pluginRand "git.yeij.top/AskaEth/Cyrene/pkg/plugins/random"
-	pluginText "git.yeij.top/AskaEth/Cyrene/pkg/plugins/text"
-	pluginWF "git.yeij.top/AskaEth/Cyrene/pkg/plugins/web_fetch"
-	pluginWS "git.yeij.top/AskaEth/Cyrene/pkg/plugins/web_search"
+	plgManager "git.yeij.top/AskaEth/Cyrene-Plugins/manager"
+	plgSDK "git.yeij.top/AskaEth/Cyrene-Plugins/sdk"
+	pluginCalc "git.yeij.top/AskaEth/Cyrene-Plugins/calculator"
+	pluginCrypto "git.yeij.top/AskaEth/Cyrene-Plugins/crypto"
+	pluginDate "git.yeij.top/AskaEth/Cyrene-Plugins/datetime"
+	pluginFile "git.yeij.top/AskaEth/Cyrene-Plugins/file"
+	pluginHTTP "git.yeij.top/AskaEth/Cyrene-Plugins/http"
+	pluginJSON "git.yeij.top/AskaEth/Cyrene-Plugins/json"
+	pluginMD "git.yeij.top/AskaEth/Cyrene-Plugins/markdown"
+	pluginRand "git.yeij.top/AskaEth/Cyrene-Plugins/random"
+	pluginText "git.yeij.top/AskaEth/Cyrene-Plugins/text"
+	pluginWF "git.yeij.top/AskaEth/Cyrene-Plugins/web_fetch"
+	pluginWS "git.yeij.top/AskaEth/Cyrene-Plugins/web_search"
 )
 
 var cfg Config

backend/ai-core/go.mod

@@ -5,12 +5,16 @@ go 1.26.2
 require (
 	github.com/joho/godotenv v1.5.1
 	github.com/lib/pq v1.10.9
+	git.yeij.top/AskaEth/Cyrene/pkg/audio v0.0.0
+	git.yeij.top/AskaEth/Cyrene/pkg/dashscope v0.0.0
 	git.yeij.top/AskaEth/Cyrene/pkg/logger v0.0.0
-	git.yeij.top/AskaEth/Cyrene/pkg/plugins v0.0.0
+	git.yeij.top/AskaEth/Cyrene-Plugins v0.0.0
 	gopkg.in/yaml.v3 v3.0.1
 )
 
 replace (
+	git.yeij.top/AskaEth/Cyrene/pkg/audio => ../pkg/audio
+	git.yeij.top/AskaEth/Cyrene/pkg/dashscope => ../pkg/dashscope
 	git.yeij.top/AskaEth/Cyrene/pkg/logger => ../pkg/logger
-	git.yeij.top/AskaEth/Cyrene/pkg/plugins => ../pkg/plugins
+	git.yeij.top/AskaEth/Cyrene-Plugins => ../cyrene-plugins
 )

backend/ai-core/internal/background/thinker.go

@@ -18,8 +18,8 @@ import (
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/persona"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/tools"
 
-	plgManager "git.yeij.top/AskaEth/Cyrene/pkg/plugins/manager"
-	plgSDK "git.yeij.top/AskaEth/Cyrene/pkg/plugins/sdk"
+	plgManager "git.yeij.top/AskaEth/Cyrene-Plugins/manager"
+	plgSDK "git.yeij.top/AskaEth/Cyrene-Plugins/sdk"
 )
 
 // PendingThought 待推送的后台思考

backend/ai-core/internal/llm/asr.go

@@ -1,31 +1,30 @@
 package llm
 
 import (
-	"bytes"
 	"context"
-	"encoding/base64"
-	"encoding/json"
 	"fmt"
 	"io"
 	"net/http"
 	"net/url"
 	"strings"
 	"time"
+
+	"git.yeij.top/AskaEth/Cyrene/pkg/audio"
+	"git.yeij.top/AskaEth/Cyrene/pkg/dashscope"
 )
 
 // ASRProvider handles speech-to-text transcription.
 type ASRProvider interface {
-	Transcribe(ctx context.Context, audioURL string) (string, error)
+	Transcribe(ctx context.Context, audioURL, language string) (string, error)
 	IsAvailable() bool
 	ModelName() string
 }
 
 // DashScopeASRProvider uses DashScope Paraformer API for offline speech recognition.
 type DashScopeASRProvider struct {
-	apiKey  string
-	baseURL string
-	model   string
-	client  *http.Client
+	model  string
+	client *dashscope.RESTClient
+	http   *http.Client
 }
 
 // NewDashScopeASRProvider creates a DashScope ASR provider.
@@ -34,16 +33,15 @@ func NewDashScopeASRProvider(baseURL, apiKey, model string) *DashScopeASRProvide
 		model = "qwen3-asr-flash-2026-02-10"
 	}
 	return &DashScopeASRProvider{
-		apiKey:  apiKey,
-		baseURL: baseURL,
-		model:   model,
-		client:  &http.Client{Timeout: 60 * time.Second},
+		model:  model,
+		client: dashscope.NewRESTClient(apiKey),
+		http:   &http.Client{Timeout: 60 * time.Second},
 	}
 }
 
 // IsAvailable returns true if the API key is configured.
 func (p *DashScopeASRProvider) IsAvailable() bool {
-	return p.apiKey != ""
+	return p.client.IsAvailable()
 }
 
 // ModelName returns the ASR model name.
@@ -51,34 +49,6 @@ func (p *DashScopeASRProvider) ModelName() string {
 	return p.model
 }
 
-type asrRequest struct {
-	Model      string       `json:"model"`
-	Input      asrInput     `json:"input"`
-	Parameters asrParams    `json:"parameters"`
-}
-
-type asrInput struct {
-	Audio string `json:"audio"`
-}
-
-type asrParams struct {
-	Format     string `json:"format,omitempty"`
-	SampleRate int    `json:"sample_rate,omitempty"`
-	Language   string `json:"language,omitempty"`
-}
-
-type asrResponse struct {
-	Output struct {
-		Text string `json:"text"`
-	} `json:"output"`
-	Usage struct {
-		TotalTokens int `json:"total_tokens"`
-	} `json:"usage"`
-	RequestID string `json:"request_id"`
-	Code      string `json:"code,omitempty"`
-	Message   string `json:"message,omitempty"`
-}
-
 // downloadAudio fetches audio data from a URL and returns the bytes with inferred format.
 func (p *DashScopeASRProvider) downloadAudio(ctx context.Context, audioURL string) ([]byte, string, error) {
 	req, err := http.NewRequestWithContext(ctx, "GET", audioURL, nil)
@@ -86,7 +56,7 @@ func (p *DashScopeASRProvider) downloadAudio(ctx context.Context, audioURL strin
 		return nil, "", fmt.Errorf("create download request: %w", err)
 	}
 
-	resp, err := p.client.Do(req)
+	resp, err := p.http.Do(req)
 	if err != nil {
 		return nil, "", fmt.Errorf("download failed: %w", err)
 	}
@@ -103,7 +73,6 @@ func (p *DashScopeASRProvider) downloadAudio(ctx context.Context, audioURL strin
 
 // inferAudioFormat determines the audio format from URL extension or Content-Type header.
 func inferAudioFormat(urlStr, contentType string) string {
-	// Try URL extension first
 	u, err := url.Parse(urlStr)
 	if err == nil {
 		path := u.Path
@@ -115,7 +84,6 @@ func inferAudioFormat(urlStr, contentType string) string {
 			}
 		}
 	}
-	// Fallback: use Content-Type
 	if strings.Contains(contentType, "audio/amr") || strings.Contains(contentType, "amr") {
 		return "amr"
 	}
@@ -130,14 +98,8 @@ func inferAudioFormat(urlStr, contentType string) string {
 	}
 	return "amr" // default for QQ voice messages
 }
-// asrEndpoint derives the DashScope ASR REST endpoint from the provider base URL.
-func asrEndpoint(baseURL string) string {
-	if u, err := url.Parse(baseURL); err == nil {
-		return fmt.Sprintf("%s://%s/api/v1/services/audio/asr/asr", u.Scheme, u.Host)
-	}
-	return strings.TrimRight(baseURL, "/") + "/api/v1/services/audio/asr/asr"
-}
-func (p *DashScopeASRProvider) Transcribe(ctx context.Context, audioURL string) (string, error) {
+
+func (p *DashScopeASRProvider) Transcribe(ctx context.Context, audioURL, language string) (string, error) {
 	if !p.IsAvailable() {
 		return "", fmt.Errorf("DashScope ASR API key not configured")
 	}
@@ -147,50 +109,15 @@ func (p *DashScopeASRProvider) Transcribe(ctx context.Context, audioURL string)
 		return "", fmt.Errorf("download audio: %w", err)
 	}
 
-	audioB64 := base64.StdEncoding.EncodeToString(audioData)
-
-	reqBody := asrRequest{
-		Model: p.model,
-		Input: asrInput{
-			Audio: fmt.Sprintf("data:audio/%s;base64,%s", format, audioB64),
-		},
-		Parameters: asrParams{
-			Format:   format,
-			Language: "zh",
-		},
-	}
-	bodyBytes, err := json.Marshal(reqBody)
+	// 转码为 16kHz mono PCM，提升识别兼容性
+	pcmData, err := audio.ConvertToPCM16(audioData, format)
 	if err != nil {
-		return "", fmt.Errorf("marshal ASR request: %w", err)
+		return "", fmt.Errorf("audio transcode: %w", err)
 	}
 
-	asrURL := asrEndpoint(p.baseURL)
-	req, err := http.NewRequestWithContext(ctx, "POST", asrURL, bytes.NewReader(bodyBytes))
-	if err != nil {
-		return "", fmt.Errorf("create ASR request: %w", err)
-	}
-	req.Header.Set("Content-Type", "application/json")
-	req.Header.Set("Authorization", "Bearer "+p.apiKey)
-
-	resp, err := p.client.Do(req)
-	if err != nil {
-		return "", fmt.Errorf("ASR request failed: %w", err)
-	}
-	defer resp.Body.Close()
-
-	respBytes, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return "", fmt.Errorf("read ASR response: %w", err)
+	if language == "" || language == "auto" {
+		language = "zh"
 	}
 
-	var asrResp asrResponse
-	if err := json.Unmarshal(respBytes, &asrResp); err != nil {
-		return "", fmt.Errorf("parse ASR response: %w", err)
-	}
-
-	if asrResp.Code != "" && asrResp.Code != "0" {
-		return "", fmt.Errorf("ASR error: %s (code=%s)", asrResp.Message, asrResp.Code)
-	}
-
-	return asrResp.Output.Text, nil
+	return p.client.Transcribe(ctx, p.model, pcmData, "pcm", 16000, language)
 }

backend/ai-core/internal/orchestrator/orchestrator.go

@@ -19,7 +19,7 @@ import (
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/bus"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/scheduler"
 
-	plgManager "git.yeij.top/AskaEth/Cyrene/pkg/plugins/manager"
+	plgManager "git.yeij.top/AskaEth/Cyrene-Plugins/manager"
 )
 
 // Orchestrator 对话编排器 v2.0
@@ -878,7 +878,7 @@ func (o *Orchestrator) preprocessVoice(ctx context.Context, message string, voic
 
 	var transcriptions []string
 	for i, url := range voiceURLs {
-		text, err := o.asrProvider.Transcribe(ctx, url)
+		text, err := o.asrProvider.Transcribe(ctx, url, "zh")
 		if err != nil {
 			logger.Printf("[orchestrator] 语音 %d 转录失败: %v", i, err)
 			continue

backend/ai-core/internal/orchestrator/synthesizer.go

@@ -10,8 +10,8 @@ import (
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/llm"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/model"
 	"git.yeij.top/AskaEth/Cyrene/pkg/logger"
-	plgManager "git.yeij.top/AskaEth/Cyrene/pkg/plugins/manager"
-	plgSDK "git.yeij.top/AskaEth/Cyrene/pkg/plugins/sdk"
+	plgManager "git.yeij.top/AskaEth/Cyrene-Plugins/manager"
+	plgSDK "git.yeij.top/AskaEth/Cyrene-Plugins/sdk"
 )
 
 // Synthesizer 主会话综合器

backend/ai-core/internal/tools/plugin_manager_client.go

@@ -1,128 +0,0 @@
-package tools
-
-import (
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"net/http"
-	"time"
-)
-
-// PluginManagerClient calls the plugin-manager service.
-type PluginManagerClient struct {
-	baseURL    string
-	httpClient *http.Client
-}
-
-// PMToolDefinition matches the plugin-manager tool definition format.
-type PMToolDefinition struct {
-	ID          string                 `json:"id"`
-	Name        string                 `json:"name"`
-	DisplayName string                 `json:"displayName"`
-	Description string                 `json:"description"`
-	Category    string                 `json:"category"`
-	Complexity  string                 `json:"complexity"`
-	Parameters  map[string]interface{} `json:"parameters"`
-	DangerLevel string                 `json:"danger_level,omitempty"`
-}
-
-// PMToolResult matches the plugin-manager execution result.
-type PMToolResult struct {
-	ToolName string `json:"tool_name"`
-	Success  bool   `json:"success"`
-	Output   string `json:"output,omitempty"`
-	Error    string `json:"error,omitempty"`
-}
-
-// PMPluginInfo matches plugin-manager plugin info.
-type PMPluginInfo struct {
-	Name    string   `json:"name"`
-	Version string   `json:"version"`
-	Status  string   `json:"status"`
-	Enabled bool     `json:"enabled"`
-	Tools   []string `json:"tools"`
-}
-
-func NewPluginManagerClient(baseURL string) *PluginManagerClient {
-	return &PluginManagerClient{
-		baseURL:    baseURL,
-		httpClient: &http.Client{Timeout: 10 * time.Second},
-	}
-}
-
-// GetToolDefinitions fetches all tool definitions from plugin-manager.
-func (c *PluginManagerClient) GetToolDefinitions(ctx context.Context) ([]PMToolDefinition, error) {
-	req, _ := http.NewRequestWithContext(ctx, "GET", c.baseURL+"/api/v1/tools", nil)
-	resp, err := c.httpClient.Do(req)
-	if err != nil {
-		return nil, fmt.Errorf("plugin-manager GetToolDefinitions: %w", err)
-	}
-	defer resp.Body.Close()
-
-	var body struct {
-		Tools []PMToolDefinition `json:"tools"`
-	}
-	if err := json.NewDecoder(resp.Body).Decode(&body); err != nil {
-		return nil, fmt.Errorf("plugin-manager decode tools: %w", err)
-	}
-	return body.Tools, nil
-}
-
-// ExecuteTool calls a tool on plugin-manager by ID.
-func (c *PluginManagerClient) ExecuteTool(ctx context.Context, toolID string, args map[string]interface{}) (*PMToolResult, error) {
-	body, _ := json.Marshal(map[string]interface{}{"arguments": args})
-	url := fmt.Sprintf("%s/api/v1/tools/%s/execute", c.baseURL, toolID)
-	req, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
-	if err != nil {
-		return nil, err
-	}
-	req.Header.Set("Content-Type", "application/json")
-
-	resp, err := c.httpClient.Do(req)
-	if err != nil {
-		return nil, fmt.Errorf("plugin-manager ExecuteTool: %w", err)
-	}
-	defer resp.Body.Close()
-
-	var result PMToolResult
-	if err := json.NewDecoder(resp.Body).Decode(&result); err != nil {
-		return nil, fmt.Errorf("plugin-manager decode result: %w", err)
-	}
-	return &result, nil
-}
-
-// ListPlugins fetches all installed plugins from plugin-manager.
-func (c *PluginManagerClient) ListPlugins(ctx context.Context) ([]PMPluginInfo, error) {
-	req, _ := http.NewRequestWithContext(ctx, "GET", c.baseURL+"/api/v1/plugins", nil)
-	resp, err := c.httpClient.Do(req)
-	if err != nil {
-		return nil, err
-	}
-	defer resp.Body.Close()
-
-	var body struct {
-		Plugins []PMPluginInfo `json:"plugins"`
-	}
-	if err := json.NewDecoder(resp.Body).Decode(&body); err != nil {
-		return nil, err
-	}
-	return body.Plugins, nil
-}
-
-// AdaptDefinitions converts PM tool definitions to ai-core ToolDefinition format.
-func (c *PluginManagerClient) AdaptDefinitions(ctx context.Context) ([]ToolDefinition, error) {
-	pmDefs, err := c.GetToolDefinitions(ctx)
-	if err != nil {
-		return nil, err
-	}
-	defs := make([]ToolDefinition, 0, len(pmDefs))
-	for _, d := range pmDefs {
-		defs = append(defs, ToolDefinition{
-			Name:        d.Name,
-			Description: d.Description,
-			Parameters:  d.Parameters,
-		})
-	}
-	return defs, nil
-}