fix(tts): restore Edge and DashScope read-aloud (#646)

mateclaw-server/pom.xml

@@ -53,6 +53,20 @@
             </exclusions>
         </dependency>
 
+        <!-- CosyVoice uses the native WebSocket API, not OpenAI audio/speech. -->
+        <dependency>
+            <groupId>com.alibaba</groupId>
+            <artifactId>dashscope-sdk-java</artifactId>
+            <version>2.22.2</version>
+            <exclusions>
+                <!-- Spring Boot already provides Logback. -->
+                <exclusion>
+                    <groupId>org.slf4j</groupId>
+                    <artifactId>slf4j-simple</artifactId>
+                </exclusion>
+            </exclusions>
+        </dependency>
+
         <!-- ===== Spring AI Alibaba Graph Core (StateGraph workflow engine) ===== -->
         <dependency>
             <groupId>com.alibaba.cloud.ai</groupId>

mateclaw-server/src/main/java/vip/mate/tts/provider/DashScopeTtsProvider.java

@@ -1,9 +1,9 @@
 package vip.mate.tts.provider;
 
-import cn.hutool.http.HttpRequest;
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisParam;
-import cn.hutool.http.HttpResponse;
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisAudioFormat;
-import com.fasterxml.jackson.databind.ObjectMapper;
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesizer;
-import com.fasterxml.jackson.databind.node.ObjectNode;
+import java.nio.ByteBuffer;
 import lombok.RequiredArgsConstructor;
 import lombok.extern.slf4j.Slf4j;
 import org.springframework.stereotype.Component;
@@ -17,11 +17,11 @@ import vip.mate.tts.TtsResult;
 import java.util.List;
 
 /**
- * DashScope TTS Provider — 使用 CosyVoice（OpenAI 兼容接口）
+ * DashScope TTS Provider — 使用 CosyVoice 原生 WebSocket SDK
  * <p>
  * 同步模式，直接返回音频流。
  * 复用已有的 DashScope LLM provider 的 API Key。
- * API 文档: https://help.aliyun.com/zh/model-studio/developer-reference/cosyvoice-openai-compatible
+ * API 文档: https://help.aliyun.com/zh/model-studio/developer-reference/cosyvoice-java-sdk
  *
  * @author MateClaw Team
  */
@@ -31,11 +31,9 @@ import java.util.List;
 public class DashScopeTtsProvider implements TtsProvider {
 
     private final ModelProviderService modelProviderService;
-    private final ObjectMapper objectMapper;
 
-    private static final String BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1";
     private static final String DEFAULT_MODEL = "cosyvoice-v2";
-    private static final String DEFAULT_VOICE = "longxiaochun";
+    private static final String DEFAULT_VOICE = "longxiaochun_v2";
 
     @Override
     public String id() {
@@ -69,9 +67,8 @@ public class DashScopeTtsProvider implements TtsProvider {
     @Override
     public List<String> availableVoices() {
         return List.of(
-                "longxiaochun", "longxiaoxia", "longlaotie", "longshu",
+                "longxiaochun_v2", "longxiaoxia_v2", "longshu_v2",
-                "longhua", "longshuo", "longjielidou", "longmiao",
+                "longhua_v2", "longwan_v2", "longcheng_v2"
-                "longyue", "longfei", "longtong", "longxiang"
         );
     }
 
@@ -83,7 +80,7 @@ public class DashScopeTtsProvider implements TtsProvider {
     @Override
     public TtsResult synthesize(TtsRequest request, SystemSettingsDTO config) {
         String apiKey = getDashScopeApiKey();
-        if (apiKey == null) {
+        if (apiKey == null || apiKey.isBlank()) {
             return TtsResult.failure("DashScope API Key 未配置");
         }
 
@@ -93,36 +90,33 @@ public class DashScopeTtsProvider implements TtsProvider {
             String voice = request.getVoice() != null && !request.getVoice().isBlank()
                     ? request.getVoice() : DEFAULT_VOICE;
 
-            ObjectNode body = objectMapper.createObjectNode();
+            if ((request.getVoice() == null || request.getVoice().isBlank())
-            body.put("model", model);
+                    && availableVoices().contains(config.getTtsDefaultVoice() == null ? "" : config.getTtsDefaultVoice())) {
-            body.put("input", request.getText());
+                voice = config.getTtsDefaultVoice();
-            body.put("voice", voice);
-            body.put("response_format", "mp3");
-            if (request.getSpeed() != null && request.getSpeed() != 1.0) {
-                body.put("speed", request.getSpeed());
             }
-
+            SpeechSynthesisParam param = SpeechSynthesisParam.builder()
-            String endpoint = BASE_URL + "/audio/speech";
+                    .apiKey(apiKey)
-            HttpResponse response = HttpRequest.post(endpoint)
+                    .model(model)
-                    .header("Authorization", "Bearer " + apiKey)
+                    .voice(voice)
-                    .header("Content-Type", "application/json")
+                    .format(SpeechSynthesisAudioFormat.MP3_24000HZ_MONO_256KBPS)
-                    .body(body.toString())
+                    .speechRate(request.getSpeed() == null ? 1.0f : request.getSpeed().floatValue())
-                    .timeout(60_000)
+                    .build();
-                    .execute();
+            SpeechSynthesizer synthesizer = new SpeechSynthesizer(param, null);
-
+            try {
-            if (response.getStatus() == 200) {
+                ByteBuffer audio = synthesizer.call(request.getText(), 60_000);
-                byte[] audioData = response.bodyBytes();
+                if (audio == null || !audio.hasRemaining()) {
+                    return TtsResult.failure("DashScope TTS 未返回音频数据");
+                }
+                byte[] audioData = new byte[audio.remaining()];
+                audio.get(audioData);
                 log.info("[DashScope TTS] Synthesized {} bytes (model={}, voice={})", audioData.length, model, voice);
                 return TtsResult.success(audioData, "audio/mpeg", "mp3");
-            } else {
+            } finally {
-                String errBody = response.body();
+                synthesizer.getDuplexApi().close(1000, "bye");
-                log.warn("[DashScope TTS] Failed: HTTP {} - {}", response.getStatus(), errBody);
-                return TtsResult.failure(TtsResponseDiagnostics.failureMessage(
-                        "DashScope TTS", endpoint, response.getStatus(), errBody));
             }
         } catch (Exception e) {
             log.error("[DashScope TTS] Error: {}", e.getMessage(), e);
-            return TtsResult.failure("DashScope TTS 异常: " + e.getMessage());
+            return TtsResult.failure("DashScope TTS 异常: " + TtsResponseDiagnostics.snippet(e.getMessage()));
         }
     }
 

mateclaw-server/src/main/java/vip/mate/tts/provider/EdgeTtsProvider.java

@@ -13,6 +13,13 @@ import java.net.http.HttpClient;
 import java.net.http.WebSocket;
 import java.nio.ByteBuffer;
 import java.time.Duration;
+import java.time.Instant;
+import java.time.ZoneOffset;
+import java.time.format.DateTimeFormatter;
+import java.nio.charset.StandardCharsets;
+import java.security.MessageDigest;
+import java.util.HexFormat;
+import java.util.Locale;
 import java.util.List;
 import java.util.UUID;
 import java.util.concurrent.CompletableFuture;
@@ -31,8 +38,9 @@ import java.util.concurrent.TimeUnit;
 @Component
 public class EdgeTtsProvider implements TtsProvider {
 
-    private static final String WS_URL = "wss://speech.platform.bing.com/consumer/speech/synthesize/readaloud";
+    private static final String WS_URL = "wss://speech.platform.bing.com/consumer/speech/synthesize/readaloud/edge/v1";
     private static final String TRUSTED_CLIENT_TOKEN = "6A5AA1D4EAFF4E9FB37E23D68491D6F4";
+    private static final String EDGE_VERSION = "143.0.3650.75";
     private static final String DEFAULT_VOICE_ZH = "zh-CN-XiaoxiaoNeural";
     private static final String DEFAULT_VOICE_EN = "en-US-MichelleNeural";
     private static final String OUTPUT_FORMAT = "audio-24khz-48kbitrate-mono-mp3";
@@ -99,7 +107,9 @@ public class EdgeTtsProvider implements TtsProvider {
     private byte[] synthesizeViaWebSocket(String text, String voice, String rate) throws Exception {
         String requestId = UUID.randomUUID().toString().replace("-", "");
         String wsUrl = WS_URL + "?TrustedClientToken=" + TRUSTED_CLIENT_TOKEN
-                + "&ConnectionId=" + requestId;
+                + "&ConnectionId=" + requestId
+                + "&Sec-MS-GEC=" + securityToken(Instant.now())
+                + "&Sec-MS-GEC-Version=1-" + EDGE_VERSION;
 
         ByteArrayOutputStream audioBuffer = new ByteArrayOutputStream();
         CompletableFuture<byte[]> resultFuture = new CompletableFuture<>();
@@ -110,9 +120,13 @@ public class EdgeTtsProvider implements TtsProvider {
 
         WebSocket ws = client.newWebSocketBuilder()
                 .header("Origin", "chrome-extension://jdiccldimpdaibmpdkjnbmckianbfold")
-                .header("User-Agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36")
+                .connectTimeout(Duration.ofSeconds(10))
+                .header("User-Agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"
+                        + " (KHTML, like Gecko) Chrome/143.0.0.0 Safari/537.36 Edg/143.0.0.0")
+                .header("Cookie", "muid=" + UUID.randomUUID().toString().replace("-", "").toUpperCase(Locale.ROOT) + ";")
                 .buildAsync(URI.create(wsUrl), new WebSocket.Listener() {
                     private final StringBuilder textBuffer = new StringBuilder();
+                    private final ByteArrayOutputStream binaryBuffer = new ByteArrayOutputStream();
 
                     @Override
                     public void onOpen(WebSocket webSocket) {
@@ -135,13 +149,21 @@ public class EdgeTtsProvider implements TtsProvider {
 
                     @Override
                     public CompletionStage<?> onBinary(WebSocket webSocket, ByteBuffer data, boolean last) {
-                        // 二进制帧：前 2 字节是 header 长度（大端），跳过 header
+                        byte[] fragment = new byte[data.remaining()];
-                        byte[] bytes = new byte[data.remaining()];
+                        data.get(fragment);
-                        data.get(bytes);
+                        binaryBuffer.writeBytes(fragment);
-                        // 查找 "Path:audio\r\n" 后的音频数据
+                        if (last) {
-                        int headerEnd = findHeaderEnd(bytes);
+                            byte[] bytes = binaryBuffer.toByteArray();
-                        if (headerEnd >= 0 && headerEnd < bytes.length) {
+                            binaryBuffer.reset();
-                            audioBuffer.write(bytes, headerEnd, bytes.length - headerEnd);
+                            int headerEnd = findHeaderEnd(bytes);
+                            if (headerEnd >= 2 && headerEnd <= bytes.length) {
+                                String headers = new String(bytes, 2, headerEnd - 2, StandardCharsets.UTF_8);
+                                if (headers.contains("Path:audio\r\n")) {
+                                    audioBuffer.write(bytes, headerEnd, bytes.length - headerEnd);
+                                }
+                            } else {
+                                resultFuture.completeExceptionally(new IllegalStateException("Invalid Edge TTS audio frame"));
+                            }
                         }
                         webSocket.request(1);
                         return null;
@@ -150,7 +172,8 @@ public class EdgeTtsProvider implements TtsProvider {
                     @Override
                     public CompletionStage<?> onClose(WebSocket webSocket, int statusCode, String reason) {
                         if (!resultFuture.isDone()) {
-                            resultFuture.complete(audioBuffer.toByteArray());
+                            resultFuture.completeExceptionally(new IllegalStateException(
+                                    "Edge TTS connection closed before turn.end: " + statusCode));
                         }
                         return null;
                     }
@@ -163,29 +186,46 @@ public class EdgeTtsProvider implements TtsProvider {
                     }
                 }).join();
 
-        // 发送配置消息
+        try {
-        String configMsg = "Content-Type:application/json; charset=utf-8\r\n"
+            String timestamp = DateTimeFormatter.ofPattern(
-                + "Path:speech.config\r\n\r\n"
+                    "EEE MMM dd yyyy HH:mm:ss 'GMT+0000 (Coordinated Universal Time)'", Locale.US)
-                + "{\"context\":{\"synthesis\":{\"audio\":{\"metadataoptions\":{"
+                    .withZone(ZoneOffset.UTC).format(Instant.now());
-                + "\"sentenceBoundaryEnabled\":false,\"wordBoundaryEnabled\":false},"
+            // 发送配置消息
-                + "\"outputFormat\":\"" + OUTPUT_FORMAT + "\"}}}}\r\n";
+            String configMsg = "X-Timestamp:" + timestamp + "\r\nContent-Type:application/json; charset=utf-8\r\n"
-        ws.sendText(configMsg, true);
+                    + "Path:speech.config\r\n\r\n"
+                    + "{\"context\":{\"synthesis\":{\"audio\":{\"metadataoptions\":{"
+                    + "\"sentenceBoundaryEnabled\":false,\"wordBoundaryEnabled\":false},"
+                    + "\"outputFormat\":\"" + OUTPUT_FORMAT + "\"}}}}\r\n";
+            ws.sendText(configMsg, true).join();
 
-        // 发送 SSML
+            // 发送 SSML
-        String escapedText = escapeXml(text);
+            String escapedText = escapeXml(text);
-        String ssml = "<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='en-US'>"
+            String ssml = "<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='en-US'>"
-                + "<voice name='" + voice + "'>"
+                    + "<voice name='" + voice + "'>"
-                + "<prosody pitch='+0Hz' rate='" + rate + "' volume='+0%'>"
+                    + "<prosody pitch='+0Hz' rate='" + rate + "' volume='+0%'>"
-                + escapedText
+                    + escapedText
-                + "</prosody></voice></speak>";
+                    + "</prosody></voice></speak>";
 
-        String ssmlMsg = "X-RequestId:" + requestId + "\r\n"
+            String ssmlMsg = "X-RequestId:" + requestId + "\r\n"
-                + "Content-Type:application/ssml+xml\r\n"
+                    + "X-Timestamp:" + timestamp + "Z\r\n"
-                + "Path:ssml\r\n\r\n" + ssml;
+                    + "Content-Type:application/ssml+xml\r\n"
-        ws.sendText(ssmlMsg, true);
+                    + "Path:ssml\r\n\r\n" + ssml;
+            ws.sendText(ssmlMsg, true).join();
 
-        // 等待结果（最多 60 秒）
+            // 等待结果（最多 60 秒）
-        return resultFuture.get(60, TimeUnit.SECONDS);
+            return resultFuture.get(60, TimeUnit.SECONDS);
+        } finally {
+            ws.abort();
+        }
+    }
+
+    // Edge validates a SHA-256 of the five-minute Windows FILETIME bucket and client token.
+    static String securityToken(Instant now) throws Exception {
+        long seconds = now.getEpochSecond() + 11_644_473_600L;
+        long ticks = (seconds - seconds % 300) * 10_000_000L;
+        byte[] digest = MessageDigest.getInstance("SHA-256")
+                .digest((ticks + TRUSTED_CLIENT_TOKEN).getBytes(StandardCharsets.US_ASCII));
+        return HexFormat.of().withUpperCase().formatHex(digest);
     }
 
     private int findHeaderEnd(byte[] data) {
@@ -200,7 +240,7 @@ public class EdgeTtsProvider implements TtsProvider {
             return requestedVoice;
         }
         String configVoice = config.getTtsDefaultVoice();
-        if (configVoice != null && !configVoice.isBlank()) {
+        if (configVoice != null && availableVoices().contains(configVoice)) {
             return configVoice;
         }
         // 自动语言检测：CJK 字符比例 > 30% 使用中文语音

mateclaw-server/src/test/java/vip/mate/tts/provider/DashScopeTtsProviderTest.java

@@ -0,0 +1,78 @@
+package vip.mate.tts.provider;
+
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesizer;
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisParam;
+import com.alibaba.dashscope.audio.ttsv2.SpeechSynthesisAudioFormat;
+import org.junit.jupiter.api.Test;
+import vip.mate.llm.service.ModelProviderService;
+import vip.mate.system.model.SystemSettingsDTO;
+import vip.mate.tts.TtsRequest;
+
+import java.nio.ByteBuffer;
+import java.util.concurrent.atomic.AtomicReference;
+
+import static org.junit.jupiter.api.Assertions.*;
+import static org.mockito.Mockito.*;
+
+class DashScopeTtsProviderTest {
+    @Test
+    void defaultVoiceMatchesCosyVoiceV2() {
+        var provider = new DashScopeTtsProvider(mock(ModelProviderService.class));
+        assertEquals("longxiaochun_v2", provider.defaultVoice());
+        assertTrue(provider.availableVoices().contains(provider.defaultVoice()));
+    }
+
+    @Test
+    void usesNativeSdkAndClosesConnection() {
+        var service = mock(ModelProviderService.class, RETURNS_DEEP_STUBS);
+        when(service.getProviderConfig("dashscope").getApiKey()).thenReturn("test-key");
+        AtomicReference<SpeechSynthesisParam> parameters = new AtomicReference<>();
+        try (var synths = mockConstruction(SpeechSynthesizer.class,
+                withSettings().defaultAnswer(RETURNS_DEEP_STUBS), (synth, context) -> {
+                    parameters.set((SpeechSynthesisParam) context.arguments().get(0));
+                    doReturn(ByteBuffer.wrap(new byte[]{1, 2, 3})).when(synth).call("Hello", 60000);
+                })) {
+            // A default voice from another provider must not break DashScope fallback.
+            var config = new SystemSettingsDTO();
+            config.setTtsDefaultVoice("zh-CN-XiaoxiaoNeural");
+            var result = new DashScopeTtsProvider(service).synthesize(
+                    TtsRequest.builder().text("Hello").speed(1.5).build(), config);
+            assertTrue(result.isSuccess(), result.getErrorMessage());
+            assertArrayEquals(new byte[]{1, 2, 3}, result.getAudioData());
+            assertEquals("cosyvoice-v2", parameters.get().getModel());
+            assertEquals("longxiaochun_v2", parameters.get().getVoice());
+            assertEquals(1.5f, parameters.get().getSpeechRate());
+            assertEquals(SpeechSynthesisAudioFormat.MP3_24000HZ_MONO_256KBPS, parameters.get().getFormat());
+            verify(synths.constructed().getFirst().getDuplexApi()).close(1000, "bye");
+        }
+    }
+    @Test
+    void closesConnectionWhenSynthesisFails() {
+        var service = mock(ModelProviderService.class, RETURNS_DEEP_STUBS);
+        when(service.getProviderConfig("dashscope").getApiKey()).thenReturn("test-key");
+        try (var synths = mockConstruction(SpeechSynthesizer.class,
+                withSettings().defaultAnswer(RETURNS_DEEP_STUBS), (synth, context) ->
+                    doThrow(new IllegalStateException("invalid API key")).when(synth).call("Hello", 60000))) {
+            var result = new DashScopeTtsProvider(service).synthesize(
+                    TtsRequest.builder().text("Hello").build(), new SystemSettingsDTO());
+            assertFalse(result.isSuccess());
+            assertTrue(result.getErrorMessage().contains("invalid API key"));
+            verify(synths.constructed().getFirst().getDuplexApi()).close(1000, "bye");
+        }
+    }
+
+    @Test
+    void rejectsEmptyAudio() {
+        var service = mock(ModelProviderService.class, RETURNS_DEEP_STUBS);
+        when(service.getProviderConfig("dashscope").getApiKey()).thenReturn("test-key");
+        try (var synths = mockConstruction(SpeechSynthesizer.class,
+                withSettings().defaultAnswer(RETURNS_DEEP_STUBS), (synth, context) ->
+                    doReturn(ByteBuffer.allocate(0)).when(synth).call("Hello", 60000))) {
+            var result = new DashScopeTtsProvider(service).synthesize(
+                    TtsRequest.builder().text("Hello").build(), new SystemSettingsDTO());
+            assertFalse(result.isSuccess());
+            verify(synths.constructed().getFirst().getDuplexApi()).close(1000, "bye");
+        }
+    }
+
+}

mateclaw-server/src/test/java/vip/mate/tts/provider/EdgeTtsProviderTest.java

@@ -0,0 +1,69 @@
+package vip.mate.tts.provider;
+
+import org.junit.jupiter.api.Test;
+import vip.mate.system.model.SystemSettingsDTO;
+import vip.mate.tts.TtsRequest;
+
+import java.net.URI;
+import java.net.http.HttpClient;
+import java.net.http.WebSocket;
+import java.nio.ByteBuffer;
+import java.nio.charset.StandardCharsets;
+import java.util.Arrays;
+import java.util.concurrent.CompletableFuture;
+import java.util.concurrent.atomic.AtomicReference;
+
+import static org.junit.jupiter.api.Assertions.*;
+import static org.mockito.ArgumentMatchers.*;
+import static org.mockito.Mockito.*;
+
+class EdgeTtsProviderTest {
+    @Test
+    void hashesWindowsFileTimeInFiveMinuteBuckets() throws Exception {
+        assertEquals("7ECB79D14E3AA576D2D79E6D487A1388156D91E614B1BE11C64226A29BC8DD8C",
+                EdgeTtsProvider.securityToken(java.time.Instant.EPOCH));
+        assertEquals(EdgeTtsProvider.securityToken(java.time.Instant.EPOCH),
+                EdgeTtsProvider.securityToken(java.time.Instant.ofEpochSecond(299)));
+        assertNotEquals(EdgeTtsProvider.securityToken(java.time.Instant.EPOCH),
+                EdgeTtsProvider.securityToken(java.time.Instant.ofEpochSecond(300)));
+    }
+
+    @Test
+    void usesCurrentHandshakeAndPreservesFragmentedAudio() {
+        HttpClient.Builder httpBuilder = mock(HttpClient.Builder.class, RETURNS_SELF);
+        HttpClient client = mock(HttpClient.class);
+        WebSocket.Builder wsBuilder = mock(WebSocket.Builder.class, RETURNS_SELF);
+        WebSocket socket = mock(WebSocket.class);
+        when(httpBuilder.build()).thenReturn(client);
+        when(client.newWebSocketBuilder()).thenReturn(wsBuilder);
+        AtomicReference<URI> endpoint = new AtomicReference<>();
+        AtomicReference<WebSocket.Listener> listener = new AtomicReference<>();
+        when(wsBuilder.buildAsync(any(), any())).thenAnswer(inv -> {
+            endpoint.set(inv.getArgument(0));
+            listener.set(inv.getArgument(1));
+            return CompletableFuture.completedFuture(socket);
+        });
+        when(socket.sendText(anyString(), eq(true))).thenAnswer(inv -> {
+            if (inv.<String>getArgument(0).contains("Path:ssml")) {
+                byte[] headers = "Path:audio\r\nContent-Type:audio/mpeg\r\n".getBytes(StandardCharsets.UTF_8);
+                byte[] packet = ByteBuffer.allocate(2 + headers.length + 4)
+                        .putShort((short) headers.length).put(headers).put(new byte[]{1, 2, 3, 4}).array();
+                listener.get().onBinary(socket, ByteBuffer.wrap(Arrays.copyOfRange(packet, 0, 1)), false);
+                listener.get().onBinary(socket, ByteBuffer.wrap(Arrays.copyOfRange(packet, 1, packet.length)), true);
+                listener.get().onText(socket, "Path:turn.end\r\n", true);
+            }
+            return CompletableFuture.completedFuture(socket);
+        });
+        try (var clients = mockStatic(HttpClient.class)) {
+            clients.when(HttpClient::newBuilder).thenReturn(httpBuilder);
+            var result = new EdgeTtsProvider().synthesize(
+                    TtsRequest.builder().text("Hello").build(), new SystemSettingsDTO());
+            assertEquals("/consumer/speech/synthesize/readaloud/edge/v1", endpoint.get().getPath());
+            assertTrue(endpoint.get().getQuery().matches(".*Sec-MS-GEC=[A-F0-9]{64}.*"));
+            assertTrue(endpoint.get().getQuery().contains("Sec-MS-GEC-Version=1-"));
+            assertTrue(result.isSuccess(), result.getErrorMessage());
+            assertArrayEquals(new byte[]{1, 2, 3, 4}, result.getAudioData());
+            verify(socket).abort();
+        }
+    }
+}

mateclaw-ui/src/components/chat/MessageBubble.vue

@@ -587,7 +587,7 @@ import { buildGeneratedFileNameMap, linkifyGeneratedFileUrls } from '@/utils/gen
 import { ensureModelViewer } from '@/utils/lazyModelViewer'
 import { useAuthenticatedAttachment } from '@/composables/useAuthenticatedAttachment'
 import { useToolLabel } from '@/composables/useToolLabel'
-import { http } from '@/api'
+import { http, fetchAuthenticatedBlob } from '@/api'
 import { copyToClipboard } from '@/utils/clipboard'
 import TypingCursor from './TypingCursor.vue'
 import { previewKindOf } from './preview/previewKind'
@@ -853,59 +853,59 @@ function copyMessage() {
 // --- TTS 朗读 ---
 const ttsState = ref<'idle' | 'loading' | 'playing'>('idle')
 let ttsAudio: HTMLAudioElement | null = null
+let ttsBlobUrl: string | null = null
+let ttsRequestVersion = 0
+
+function stopTts() {
+  ttsAudio?.pause()
+  if (ttsAudio) { ttsAudio.onended = null; ttsAudio.onerror = null }
+  ttsAudio = null
+  if (ttsBlobUrl) URL.revokeObjectURL(ttsBlobUrl)
+  ttsBlobUrl = null
+  ttsState.value = 'idle'
+}
 
 async function handleTts() {
   if (ttsState.value === 'playing') {
-    // 停止播放
+    stopTts()
-    ttsAudio?.pause()
-    ttsAudio = null
-    ttsState.value = 'idle'
     return
   }
-
+  if (ttsState.value === 'loading') return
   const text = displayContent.value || props.message.content || ''
-  if (!text) return
-
   const conversationId = props.message.conversationId
-  if (!conversationId) return
+  if (!text || !conversationId) return
 
+  const version = ++ttsRequestVersion
   ttsState.value = 'loading'
   try {
-    const res: any = await http.post('/tts/synthesize', {
+    // The shared interceptor already unwraps the raw TTS response body.
-      conversationId,
+    const result: any = await http.post('/tts/synthesize', { conversationId, text }, { timeout: 240_000 })
-      text,
+    if (version !== ttsRequestVersion) return
-    })
+    if (!result?.success || !result.audioUrl) {
-    if (res.data?.success && res.data?.audioUrl) {
+      throw new Error(result?.error || t('chat.ttsFailed'))
-      // 通过认证 fetch 获取音频 blob
-      const audioRes = await fetch(res.data.audioUrl, {
-        headers: { Authorization: `Bearer ${localStorage.getItem('token') || ''}` },
-      })
-      const blob = await audioRes.blob()
-      const blobUrl = URL.createObjectURL(blob)
-      ttsAudio = new Audio(blobUrl)
-      ttsAudio.onended = () => {
-        ttsState.value = 'idle'
-        URL.revokeObjectURL(blobUrl)
-        ttsAudio = null
-      }
-      ttsAudio.onerror = () => {
-        ttsState.value = 'idle'
-        URL.revokeObjectURL(blobUrl)
-        ttsAudio = null
-      }
-      ttsState.value = 'playing'
-      await ttsAudio.play()
-    } else {
-      ttsState.value = 'idle'
     }
-  } catch {
+    const blob = await fetchAuthenticatedBlob(result.audioUrl)
-    ttsState.value = 'idle'
+    if (version !== ttsRequestVersion) return
+    ttsBlobUrl = URL.createObjectURL(blob)
+    ttsAudio = new Audio(ttsBlobUrl)
+    ttsAudio.onended = stopTts
+    ttsAudio.onerror = () => {
+      stopTts()
+      mcToast.error(t('chat.ttsFailed'))
+    }
+    await ttsAudio.play()
+    if (version === ttsRequestVersion && ttsAudio) ttsState.value = 'playing'
+  } catch (error) {
+    if (version !== ttsRequestVersion) return
+    stopTts()
+    mcToast.error(error instanceof Error ? error.message : t('chat.ttsFailed'))
   }
 }
 
 onBeforeUnmount(() => {
   if (copyTimer) clearTimeout(copyTimer)
-  if (ttsAudio) { ttsAudio.pause(); ttsAudio = null }
+  ++ttsRequestVersion
+  stopTts()
   revokeAll()
 })
 

mateclaw-ui/src/components/chat/__tests__/MessageBubble.stopIndicator.test.ts

@@ -29,7 +29,7 @@ vi.mock('@/composables/useMcToast', () => ({
 vi.mock('@/composables/useToolLabel', () => ({
   useToolLabel: () => ({ getToolLabel: (name: string) => name }),
 }))
-vi.mock('@/api', () => ({ http: { get: vi.fn(), post: vi.fn() } }))
+vi.mock('@/api', () => ({ http: { get: vi.fn(), post: vi.fn() }, fetchAuthenticatedBlob: vi.fn() }))
 vi.mock('@/utils/clipboard', () => ({ copyToClipboard: vi.fn() }))
 vi.mock('@/utils/generatedFileLinks', () => ({
   buildGeneratedFileNameMap: vi.fn(() => new Map()),
@@ -58,6 +58,8 @@ vi.mock('../UserMessageContent.vue', () => ({
 }))
 
 import MessageBubble from '../MessageBubble.vue'
+import { http, fetchAuthenticatedBlob } from '@/api'
+import { mcToast } from '@/composables/useMcToast'
 
 const apps: Array<ReturnType<typeof createApp>> = []
 
@@ -92,6 +94,9 @@ function mountMessage(message: Message, locale = 'zh-CN') {
 afterEach(() => {
   apps.splice(0).forEach(app => app.unmount())
   document.body.innerHTML = ''
+  vi.restoreAllMocks()
+  vi.unstubAllGlobals()
+  vi.clearAllMocks()
 })
 
 describe('MessageBubble stop indicator', () => {
@@ -130,3 +135,69 @@ describe('MessageBubble stop indicator', () => {
     expect(host.querySelector('.stopped-indicator--interrupted')).not.toBeNull()
   })
 })
+
+
+describe('MessageBubble TTS (#646)', () => {
+  function clickReadAloud() {
+    const host = mountMessage({ id: 'tts', conversationId: 'conv', role: 'assistant',
+      content: 'Hello', contentParts: [], status: 'completed' } as Message)
+    const button = host.querySelector<HTMLButtonElement>('[title="chat.ttsPlay"]')!
+    expect(button).not.toBeNull()
+    button.click()
+    return button
+  }
+
+  it('plays the unwrapped synthesis response and releases audio on stop', async () => {
+    vi.mocked(http.post).mockResolvedValue({ success: true, audioUrl: '/api/v1/chat/files/conv/tts.mp3' } as never)
+    vi.mocked(fetchAuthenticatedBlob).mockResolvedValue(new Blob(['audio']))
+    const play = vi.fn().mockResolvedValue(undefined)
+    const pause = vi.fn()
+    vi.stubGlobal('Audio', class { play = play; pause = pause })
+    vi.spyOn(URL, 'createObjectURL').mockReturnValue('blob:tts')
+    const revoke = vi.spyOn(URL, 'revokeObjectURL').mockImplementation(() => {})
+    const button = clickReadAloud()
+    await vi.waitFor(() => expect(play).toHaveBeenCalledOnce())
+    expect(fetchAuthenticatedBlob).toHaveBeenCalledWith('/api/v1/chat/files/conv/tts.mp3')
+    button.click()
+    expect(pause).toHaveBeenCalledOnce()
+    expect(revoke).toHaveBeenCalledWith('blob:tts')
+  })
+
+  it('shows backend synthesis errors instead of silently doing nothing', async () => {
+    vi.mocked(http.post).mockResolvedValue({ success: false, error: 'DashScope: invalid voice' } as never)
+    clickReadAloud()
+    await vi.waitFor(() => expect(mcToast.error).toHaveBeenCalledWith('DashScope: invalid voice'))
+  })
+
+  it('shows audio download failures and resets the loading state', async () => {
+    vi.mocked(http.post).mockResolvedValue({ success: true, audioUrl: '/audio' } as never)
+    vi.mocked(fetchAuthenticatedBlob).mockRejectedValue(new Error('Fetch failed: 403'))
+    const button = clickReadAloud()
+    await vi.waitFor(() => expect(mcToast.error).toHaveBeenCalledWith('Fetch failed: 403'))
+    expect(button.disabled).toBe(false)
+  })
+  it('releases the audio URL when playback is rejected', async () => {
+    vi.mocked(http.post).mockResolvedValue({ success: true, audioUrl: '/audio' } as never)
+    vi.mocked(fetchAuthenticatedBlob).mockResolvedValue(new Blob(['audio']))
+    const pause = vi.fn()
+    vi.stubGlobal('Audio', class { pause = pause; play = vi.fn().mockRejectedValue(new Error('Playback blocked')) })
+    vi.spyOn(URL, 'createObjectURL').mockReturnValue('blob:rejected')
+    const revoke = vi.spyOn(URL, 'revokeObjectURL').mockImplementation(() => {})
+    clickReadAloud()
+    await vi.waitFor(() => expect(mcToast.error).toHaveBeenCalledWith('Playback blocked'))
+    expect(pause).toHaveBeenCalledOnce()
+    expect(revoke).toHaveBeenCalledWith('blob:rejected')
+  })
+
+  it('does not start playback when the component unmounts during synthesis', async () => {
+    let finish!: (value: any) => void
+    vi.mocked(http.post).mockReturnValue(new Promise(resolve => { finish = resolve }))
+    clickReadAloud()
+    apps.pop()!.unmount()
+    finish({ success: true, audioUrl: '/audio' })
+    await Promise.resolve()
+    await Promise.resolve()
+    expect(fetchAuthenticatedBlob).not.toHaveBeenCalled()
+  })
+
+})

mateclaw-ui/src/i18n/locales/en-US.ts

@@ -497,6 +497,7 @@ export default {
         },
       },
     },
+    ttsFailed: 'Speech synthesis or playback failed',
     ttsPlay: 'Read Aloud',
     ttsStop: 'Stop Reading',
     conversations: 'Conversations',

mateclaw-ui/src/i18n/locales/zh-CN.ts

@@ -497,6 +497,7 @@ export default {
         },
       },
     },
+    ttsFailed: '语音合成或播放失败',
     ttsPlay: '朗读',
     ttsStop: '停止朗读',
     conversations: '会话列表',