feat: 添加 MiMo TTS 语音合成服务

实现基于 Xiaomi MiMo TTS API 的语音合成 provider，使用冰糖音色。 - 新增 MiMoService 实现 tts.Service 接口 - 使用 chat/completions 格式，与 MiMo STT 保持一致 - 在 main.go 添加 TTS provider 切换逻辑（mimo/xiaomi） - 配置文件默认音色设为冰糖 - 包含完整测试覆盖（9 个测试用例）
2026-06-14 10:05:56 +08:00
parent b8ed49be63
commit e967d89e7e
4 changed files with 626 additions and 3 deletions
--- a/backend/internal/ai/tts/mimo.go
+++ b/backend/internal/ai/tts/mimo.go
@@ -0,0 +1,212 @@
+package tts
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"go.uber.org/zap"
+)
+
+// MiMoService 基于 Xiaomi MiMo TTS API 的语音合成实现。
+// 接口兼容 OpenAI chat/completions 格式，通过 messages 传递待合成文本与风格指令。
+type MiMoService struct {
+	apiKey   string
+	model    string
+	voice    string
+	endpoint string
+	timeout  time.Duration
+	logger   *zap.SugaredLogger
+	client   *http.Client
+}
+
+// NewMiMoService 创建 MiMo TTS 服务。
+func NewMiMoService(apiKey, model, voice, endpoint string, timeoutSec int, logger *zap.SugaredLogger) *MiMoService {
+	if model == "" {
+		model = "mimo-v2.5-tts"
+	}
+	if voice == "" {
+		voice = "冰糖"
+	}
+	if endpoint == "" {
+		endpoint = "https://api.xiaomimimo.com/v1"
+	}
+	timeout := time.Duration(timeoutSec) * time.Second
+	if timeout <= 0 {
+		timeout = 5 * time.Second
+	}
+	return &MiMoService{
+		apiKey:   apiKey,
+		model:    model,
+		voice:    voice,
+		endpoint: endpoint,
+		timeout:  timeout,
+		logger:   logger,
+		client:   &http.Client{Timeout: 30 * time.Second},
+	}
+}
+
+// mimoTTSRequest MiMo TTS API 请求体。
+type mimoTTSRequest struct {
+	Model    string           `json:"model"`
+	Messages []mimoTTSMessage `json:"messages"`
+	Audio    mimoTTSAudio     `json:"audio"`
+	Stream   bool             `json:"stream"`
+}
+
+// mimoTTSMessage MiMo TTS 消息。
+type mimoTTSMessage struct {
+	Role    string `json:"role"`    // "user"（风格指令）| "assistant"（待合成文本）
+	Content string `json:"content"`
+}
+
+// mimoTTSAudio MiMo TTS 音频配置。
+type mimoTTSAudio struct {
+	Format string `json:"format"` // "mp3" | "wav" | "pcm16"
+	Voice  string `json:"voice"`  // 预置音色 ID
+}
+
+// mimoTTSResponse MiMo TTS 非流式响应。
+type mimoTTSResponse struct {
+	Choices []struct {
+		Message struct {
+			Audio struct {
+				Data string `json:"data"` // base64 编码的音频数据
+			} `json:"audio"`
+		} `json:"message"`
+	} `json:"choices"`
+}
+
+// mimoTTSStreamResponse MiMo TTS 流式响应。
+type mimoTTSStreamResponse struct {
+	Choices []struct {
+		Delta struct {
+			Audio struct {
+				Data string `json:"data"` // base64 编码的音频数据片段
+			} `json:"audio"`
+		} `json:"delta"`
+	} `json:"choices"`
+}
+
+// SynthesizeStream 实现 tts.Service。从 textStream 读取句子，逐句调用 MiMo TTS API。
+func (m *MiMoService) SynthesizeStream(ctx context.Context, textStream <-chan string, opts Options) (<-chan Chunk, error) {
+	voice := opts.Voice
+	if voice == "" {
+		voice = m.voice
+	}
+
+	ch := make(chan Chunk, 4)
+	go func() {
+		defer close(ch)
+
+		for text := range textStream {
+			if text == "" {
+				continue
+			}
+
+			audio, err := m.synthesize(ctx, text, voice)
+			if err != nil {
+				m.logger.Warnw("mimo tts: synthesize failed", "error", err, "text", text)
+				// 静默跳过，不中断整个流
+				continue
+			}
+
+			select {
+			case ch <- Chunk{Audio: audio, IsLast: false}:
+			case <-ctx.Done():
+				return
+			}
+		}
+
+		// textStream 关闭，发送 IsLast 标记
+		select {
+		case ch <- Chunk{Audio: nil, IsLast: true}:
+		case <-ctx.Done():
+		}
+	}()
+
+	return ch, nil
+}
+
+// synthesize 调用 MiMo TTS API 合成单个句子。
+// 使用非流式调用，返回完整音频数据（base64 解码后）。
+func (m *MiMoService) synthesize(ctx context.Context, text, voice string) ([]byte, error) {
+	// 单句超时
+	ctx, cancel := context.WithTimeout(ctx, m.timeout)
+	defer cancel()
+
+	// 构建 MiMo TTS 请求：文本放在 assistant 消息中
+	reqBody := mimoTTSRequest{
+		Model: m.model,
+		Messages: []mimoTTSMessage{
+			{
+				Role:    "assistant",
+				Content: text,
+			},
+		},
+		Audio: mimoTTSAudio{
+			Format: "mp3",
+			Voice:  voice,
+		},
+		Stream: false,
+	}
+
+	payload, err := json.Marshal(reqBody)
+	if err != nil {
+		return nil, fmt.Errorf("mimo tts: marshal request: %w", err)
+	}
+
+	url := strings.TrimRight(m.endpoint, "/") + "/chat/completions"
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(payload))
+	if err != nil {
+		return nil, fmt.Errorf("mimo tts: create request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("api-key", m.apiKey)
+
+	resp, err := m.client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("mimo tts: send request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		errBody, _ := io.ReadAll(resp.Body)
+		return nil, fmt.Errorf("mimo tts: api error (status %d): %s", resp.StatusCode, string(errBody))
+	}
+
+	// 非流式响应：解析 JSON，提取 base64 音频数据
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("mimo tts: read response: %w", err)
+	}
+
+	var ttsResp mimoTTSResponse
+	if err := json.Unmarshal(respBody, &ttsResp); err != nil {
+		return nil, fmt.Errorf("mimo tts: unmarshal response: %w", err)
+	}
+
+	if len(ttsResp.Choices) == 0 {
+		return nil, fmt.Errorf("mimo tts: empty choices in response")
+	}
+
+	audioData := ttsResp.Choices[0].Message.Audio.Data
+	if audioData == "" {
+		return nil, fmt.Errorf("mimo tts: empty audio data in response")
+	}
+
+	// base64 解码音频数据
+	audio, err := base64.StdEncoding.DecodeString(audioData)
+	if err != nil {
+		return nil, fmt.Errorf("mimo tts: decode audio base64: %w", err)
+	}
+
+	return audio, nil
+}
--- a/backend/internal/ai/tts/mimo_test.go
+++ b/backend/internal/ai/tts/mimo_test.go
@@ -0,0 +1,405 @@
+package tts
+
+import (
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"go.uber.org/zap"
+)
+
+// mockMiMoTTSServer 创建模拟 MiMo TTS API 的 HTTP 服务器。
+func mockMiMoTTSServer(t *testing.T, handler http.HandlerFunc) *httptest.Server {
+	t.Helper()
+	return httptest.NewServer(handler)
+}
+
+// buildMiMoTTSResponse 构造 MiMo TTS 非流式响应 JSON。
+func buildMiMoTTSResponse(audioData string) []byte {
+	resp := mimoTTSResponse{
+		Choices: []struct {
+			Message struct {
+				Audio struct {
+					Data string `json:"data"`
+				} `json:"audio"`
+			} `json:"message"`
+		}{
+			{
+				Message: struct {
+					Audio struct {
+						Data string `json:"data"`
+					} `json:"audio"`
+				}{
+					Audio: struct {
+						Data string `json:"data"`
+					}{Data: audioData},
+				},
+			},
+		},
+	}
+	data, _ := json.Marshal(resp)
+	return data
+}
+
+func TestMiMoService_SynthesizeStream_Success(t *testing.T) {
+	var callCount int32
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&callCount, 1)
+
+		if r.Method != http.MethodPost {
+			t.Errorf("method = %s, want POST", r.Method)
+		}
+		if !strings.Contains(r.URL.Path, "/chat/completions") {
+			t.Errorf("path = %s, should contain /chat/completions", r.URL.Path)
+		}
+
+		// 验证 api-key 认证头
+		apiKey := r.Header.Get("api-key")
+		if apiKey != "test-key" {
+			t.Errorf("api-key = %q, want %q", apiKey, "test-key")
+		}
+
+		// 验证请求体
+		body, _ := io.ReadAll(r.Body)
+		var req mimoTTSRequest
+		if err := json.Unmarshal(body, &req); err != nil {
+			t.Errorf("unmarshal request: %v", err)
+		}
+		if req.Model != "mimo-v2.5-tts" {
+			t.Errorf("model = %q, want %q", req.Model, "mimo-v2.5-tts")
+		}
+		if len(req.Messages) != 1 || req.Messages[0].Role != "assistant" {
+			t.Errorf("expected 1 assistant message, got %d messages", len(req.Messages))
+		}
+		if req.Audio.Voice != "冰糖" {
+			t.Errorf("voice = %q, want %q", req.Audio.Voice, "冰糖")
+		}
+		if req.Audio.Format != "mp3" {
+			t.Errorf("format = %q, want %q", req.Audio.Format, "mp3")
+		}
+
+		// 返回假音频数据（base64 编码）
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("fake-mp3-data"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("你好", "世界", "！")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{
+		Voice: "冰糖", OutputFmt: "mp3", SampleRate: 24000,
+	})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	// 应该有 3 个音频 chunk + 1 个 IsLast 标记
+	if len(chunks) != 4 {
+		t.Fatalf("got %d chunks, want 4", len(chunks))
+	}
+
+	// 验证前 3 个有音频数据
+	for i := 0; i < 3; i++ {
+		if string(chunks[i].Audio) != "fake-mp3-data" {
+			t.Errorf("chunk[%d].Audio = %q, want %q", i, string(chunks[i].Audio), "fake-mp3-data")
+		}
+		if chunks[i].IsLast {
+			t.Errorf("chunk[%d].IsLast should be false", i)
+		}
+	}
+
+	// 验证最后一个是 IsLast
+	if !chunks[3].IsLast {
+		t.Error("last chunk should be IsLast")
+	}
+	if chunks[3].Audio != nil {
+		t.Error("last chunk Audio should be nil")
+	}
+
+	// 验证调用了 3 次 API（3 个句子）
+	if atomic.LoadInt32(&callCount) != 3 {
+		t.Errorf("API called %d times, want 3", callCount)
+	}
+}
+
+func TestMiMoService_SynthesizeStream_APIError(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		fmt.Fprintf(w, "internal error")
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("你好")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	// 应该只有一个 IsLast chunk（音频被跳过）
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	if len(chunks) != 1 {
+		t.Fatalf("got %d chunks, want 1 (IsLast only)", len(chunks))
+	}
+	if !chunks[0].IsLast {
+		t.Error("chunk should be IsLast")
+	}
+}
+
+func TestMiMoService_SynthesizeStream_Timeout(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(3 * time.Second)
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("late-mp3"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	// 1 秒超时
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 1, zap.NewNop().Sugar())
+
+	textStream := sendSentences("很长的句子")
+
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+
+	ch, err := svc.SynthesizeStream(ctx, textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	// 超时后音频被跳过，只有 IsLast
+	if len(chunks) != 1 {
+		t.Fatalf("got %d chunks, want 1", len(chunks))
+	}
+	if !chunks[0].IsLast {
+		t.Error("chunk should be IsLast")
+	}
+}
+
+func TestMiMoService_SynthesizeStream_EmptyText(t *testing.T) {
+	var callCount int32
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&callCount, 1)
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("mp3"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	// 空句子应该被跳过
+	textStream := sendSentences("", "你好", "")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	// 只有 "你好" 应该被合成
+	if atomic.LoadInt32(&callCount) != 1 {
+		t.Errorf("API called %d times, want 1", callCount)
+	}
+
+	// 1 个音频 + 1 个 IsLast
+	if len(chunks) != 2 {
+		t.Fatalf("got %d chunks, want 2", len(chunks))
+	}
+}
+
+func TestMiMoService_SynthesizeStream_ContextCancelled(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("mp3"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := make(chan string, 3)
+	textStream <- "第一句"
+	textStream <- "第二句"
+	textStream <- "第三句"
+	close(textStream)
+
+	ctx, cancel := context.WithCancel(context.Background())
+	// 立即取消
+	cancel()
+
+	ch, err := svc.SynthesizeStream(ctx, textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	// 消费 channel，应该很快结束
+	var count int
+	for range ch {
+		count++
+	}
+	// 可能收到 0 个或 1 个 chunk，取决于时序
+	t.Logf("received %d chunks after context cancel", count)
+}
+
+func TestMiMoService_SynthesizeStream_PartialFailure(t *testing.T) {
+	var callCount int32
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		n := atomic.AddInt32(&callCount, 1)
+		if n == 2 {
+			// 第二个句子失败
+			w.WriteHeader(http.StatusInternalServerError)
+			fmt.Fprintf(w, "error")
+			return
+		}
+		audioB64 := base64.StdEncoding.EncodeToString([]byte(fmt.Sprintf("mp3-%d", n)))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("第一句", "第二句", "第三句")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	// 2 个成功音频 + 1 个 IsLast（第二句被跳过）
+	if len(chunks) != 3 {
+		t.Fatalf("got %d chunks, want 3", len(chunks))
+	}
+	if !chunks[len(chunks)-1].IsLast {
+		t.Error("last chunk should be IsLast")
+	}
+}
+
+func TestMiMoService_SynthesizeStream_CustomVoice(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		body, _ := io.ReadAll(r.Body)
+		var req mimoTTSRequest
+		if err := json.Unmarshal(body, &req); err != nil {
+			t.Errorf("unmarshal request: %v", err)
+		}
+		if req.Audio.Voice != "茉莉" {
+			t.Errorf("voice = %q, want %q", req.Audio.Voice, "茉莉")
+		}
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("mp3"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("你好")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{Voice: "茉莉"})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	for range ch {
+	}
+}
+
+func TestMiMoService_SynthesizeStream_DefaultVoice(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		body, _ := io.ReadAll(r.Body)
+		var req mimoTTSRequest
+		if err := json.Unmarshal(body, &req); err != nil {
+			t.Errorf("unmarshal request: %v", err)
+		}
+		// 未指定 voice 时应使用默认 "冰糖"
+		if req.Audio.Voice != "冰糖" {
+			t.Errorf("voice = %q, want %q (default)", req.Audio.Voice, "冰糖")
+		}
+		audioB64 := base64.StdEncoding.EncodeToString([]byte("mp3"))
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(audioB64))
+	})
+	defer srv.Close()
+
+	// 不指定 voice
+	svc := NewMiMoService("test-key", "", "", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("你好")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	for range ch {
+	}
+}
+
+func TestMiMoService_SynthesizeStream_EmptyAudioData(t *testing.T) {
+	srv := mockMiMoTTSServer(t, func(w http.ResponseWriter, r *http.Request) {
+		// 返回空音频数据
+		w.Header().Set("Content-Type", "application/json")
+		w.Write(buildMiMoTTSResponse(""))
+	})
+	defer srv.Close()
+
+	svc := NewMiMoService("test-key", "", "冰糖", srv.URL, 5, zap.NewNop().Sugar())
+
+	textStream := sendSentences("你好")
+
+	ch, err := svc.SynthesizeStream(context.Background(), textStream, Options{})
+	if err != nil {
+		t.Fatalf("SynthesizeStream() error: %v", err)
+	}
+
+	var chunks []Chunk
+	for c := range ch {
+		chunks = append(chunks, c)
+	}
+
+	// 空音频数据导致错误，句子被跳过，只有 IsLast
+	if len(chunks) != 1 {
+		t.Fatalf("got %d chunks, want 1", len(chunks))
+	}
+	if !chunks[0].IsLast {
+		t.Error("chunk should be IsLast")
+	}
+}