diff --git a/app/client/api.ts b/app/client/api.ts
index f60b0e2ad..9b82959a8 100644
--- a/app/client/api.ts
+++ b/app/client/api.ts
@@ -107,7 +107,8 @@ export interface LLMModelProvider {
 
 export abstract class LLMApi {
   abstract chat(options: ChatOptions): Promise<void>;
-  abstract speech(options: SpeechOptions): Promise<ArrayBuffer>;
+  abstract speech(options: SpeechOptions): Promise<ArrayBuffer | AudioBuffer>;
+  abstract streamSpeech?(options: SpeechOptions): AsyncGenerator<AudioBuffer>;
   abstract usage(): Promise<LLMUsage>;
   abstract models(): Promise<LLMModel[]>;
 }
diff --git a/app/client/platforms/alibaba.ts b/app/client/platforms/alibaba.ts
index 19d020ddc..7427eb818 100644
--- a/app/client/platforms/alibaba.ts
+++ b/app/client/platforms/alibaba.ts
@@ -1,5 +1,10 @@
 "use client";
-import { ApiPath, Alibaba, ALIBABA_BASE_URL, REQUEST_TIMEOUT_MS } from "@/app/constant";
+import {
+  ApiPath,
+  Alibaba,
+  ALIBABA_BASE_URL,
+  REQUEST_TIMEOUT_MS,
+} from "@/app/constant";
 import {
   useAccessStore,
   useAppConfig,
@@ -89,66 +94,71 @@ export class QwenApi implements LLMApi {
     return res?.output?.choices?.at(0)?.message?.content ?? "";
   }
 
-  async speech(options: SpeechOptions): Promise<AudioBuffer> {
+  async speech(options: SpeechOptions): Promise<ArrayBuffer> {
+    throw new Error("Method not implemented.");
+  }
+
+  async *streamSpeech(options: SpeechOptions): AsyncGenerator<AudioBuffer> {
     const requestPayload = {
-        model: options.model,
-        input: {
-            text: options.input,
-            voice: options.voice,
-        },
-        speed: options.speed,
-        response_format: options.response_format,
+      model: options.model,
+      input: {
+        text: options.input,
+        voice: options.voice,
+      },
+      speed: options.speed,
+      response_format: options.response_format,
     };
     console.log("[Request] alibaba speech payload: ", requestPayload);
     const controller = new AbortController();
     options.onController?.(controller);
     try {
-        const speechPath = this.path(Alibaba.SpeechPath);
-        const speechPayload = {
-          method: "POST",
-          body: JSON.stringify(requestPayload),
-          signal: controller.signal,
-          headers: {
-            ...getHeaders(),
-            "X-DashScope-SSE": "enable",
-          },
-        };
-  
-        // make a fetch request
-        const requestTimeoutId = setTimeout(
-          () => controller.abort(),
-          REQUEST_TIMEOUT_MS,
-        );
-  
-        const res = await fetch(speechPath, speechPayload);
+      const speechPath = this.path(Alibaba.SpeechPath);
+      const speechPayload = {
+        method: "POST",
+        body: JSON.stringify(requestPayload),
+        signal: controller.signal,
+        headers: {
+          ...getHeaders(),
+          "X-DashScope-SSE": "enable",
+        },
+      };
 
-        const reader = res.body!.getReader();
-        const decoder = new TextDecoder();
-        let buffer = "";
-        let base64 = "";
-        while (true) {
-            const { done, value } = await reader.read();
-            if (done) break;
-            buffer += decoder.decode(value, { stream: true, });
-                const lines = buffer.split('\n');
-                buffer = lines.pop() || '';
+      // make a fetch request
+      const requestTimeoutId = setTimeout(
+        () => controller.abort(),
+        REQUEST_TIMEOUT_MS,
+      );
 
-                for (const line of lines) {
-                    if (line.startsWith('data:')) {
-                        const data = line.slice(5);
-                        const json = JSON.parse(data);
-                        base64 += json.output.audio.data;
-                    }
-                }
+      const res = await fetch(speechPath, speechPayload);
+
+      const reader = res.body!.getReader();
+      const decoder = new TextDecoder();
+      let buffer = "";
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) {
+          break;
+        }
+        buffer += decoder.decode(value, { stream: true });
+        const lines = buffer.split("\n");
+        buffer = lines.pop() || "";
+
+        for (const line of lines) {
+          if (line.startsWith("data:")) {
+            const data = line.slice(5);
+            const json = JSON.parse(data);
+            if (json.output.audio.data) {
+              yield this.PCMBase64ToAudioBuffer(json.output.audio.data);
+            }
+          }
         }
-        const audioBuffer = await this.PCMBase64ToAudioBuffer(base64);
-        clearTimeout(requestTimeoutId);
-        reader.releaseLock();
-        return audioBuffer;
-      } catch (e) {
-        console.log("[Request] failed to make a speech request", e);
-        throw e;
       }
+      clearTimeout(requestTimeoutId);
+      reader.releaseLock();
+    } catch (e) {
+      console.log("[Request] failed to make a speech request", e);
+      throw e;
+    }
   }
 
   async chat(options: ChatOptions) {
@@ -335,67 +345,68 @@ export class QwenApi implements LLMApi {
   // 播放 PCM base64 数据
   private async PCMBase64ToAudioBuffer(base64Data: string) {
     try {
-        // 解码 base64
-        const binaryString = atob(base64Data);
-        const bytes = new Uint8Array(binaryString.length);
-        for (let i = 0; i < binaryString.length; i++) {
-            bytes[i] = binaryString.charCodeAt(i);
-        }
+      // 解码 base64
+      const binaryString = atob(base64Data);
+      const bytes = new Uint8Array(binaryString.length);
+      for (let i = 0; i < binaryString.length; i++) {
+        bytes[i] = binaryString.charCodeAt(i);
+      }
 
-        // 转换为 AudioBuffer
-        const audioBuffer = await this.convertToAudioBuffer(bytes);
-        
-        return audioBuffer;
+      // 转换为 AudioBuffer
+      const audioBuffer = await this.convertToAudioBuffer(bytes);
+
+      return audioBuffer;
     } catch (error) {
-        console.error('播放 PCM 数据失败:', error);
-        throw error;
+      console.error("播放 PCM 数据失败:", error);
+      throw error;
     }
   }
-  
-   // 将 PCM 字节数据转换为 AudioBuffer
-   private convertToAudioBuffer(pcmData: Uint8Array) {
-    const audioContext = new (window.AudioContext || window.webkitAudioContext)();
+
+  // 将 PCM 字节数据转换为 AudioBuffer
+  private convertToAudioBuffer(pcmData: Uint8Array) {
+    const audioContext = new (window.AudioContext ||
+      window.webkitAudioContext)();
     const channels = 1;
     const sampleRate = 24000;
     return new Promise<AudioBuffer>((resolve, reject) => {
-        try {
-            let float32Array;
-            // 16位 PCM 转换为 32位浮点数
-            float32Array = this.pcm16ToFloat32(pcmData);
+      try {
+        let float32Array;
+        // 16位 PCM 转换为 32位浮点数
+        float32Array = this.pcm16ToFloat32(pcmData);
 
-            // 创建 AudioBuffer
-            const audioBuffer = audioContext.createBuffer(
-                channels,
-                float32Array.length / channels,
-                sampleRate
-            );
+        // 创建 AudioBuffer
+        const audioBuffer = audioContext.createBuffer(
+          channels,
+          float32Array.length / channels,
+          sampleRate,
+        );
 
-            // 复制数据到 AudioBuffer
-            for (let channel = 0; channel < channels; channel++) {
-                const channelData = audioBuffer.getChannelData(channel);
-                for (let i = 0; i < channelData.length; i++) {
-                    channelData[i] = float32Array[i * channels + channel];
-                }
-            }
-
-            resolve(audioBuffer);
-        } catch (error) {
-            reject(error);
+        // 复制数据到 AudioBuffer
+        for (let channel = 0; channel < channels; channel++) {
+          const channelData = audioBuffer.getChannelData(channel);
+          for (let i = 0; i < channelData.length; i++) {
+            channelData[i] = float32Array[i * channels + channel];
+          }
         }
+
+        resolve(audioBuffer);
+      } catch (error) {
+        reject(error);
+      }
     });
   }
-    // 16位 PCM 转 32位浮点数
-    private pcm16ToFloat32(pcmData: Uint8Array) {
-        const length = pcmData.length / 2;
-        const float32Array = new Float32Array(length);
-        
-        for (let i = 0; i < length; i++) {
-            const int16 = (pcmData[i * 2 + 1] << 8) | pcmData[i * 2];
-            const int16Signed = int16 > 32767 ? int16 - 65536 : int16;
-            float32Array[i] = int16Signed / 32768;
-        }
-        
-        return float32Array;
+  // 16位 PCM 转 32位浮点数
+  private pcm16ToFloat32(pcmData: Uint8Array) {
+    const length = pcmData.length / 2;
+    const float32Array = new Float32Array(length);
+
+    for (let i = 0; i < length; i++) {
+      const int16 = (pcmData[i * 2 + 1] << 8) | pcmData[i * 2];
+      const int16Signed = int16 > 32767 ? int16 - 65536 : int16;
+      float32Array[i] = int16Signed / 32768;
     }
+
+    return float32Array;
+  }
 }
-export { Alibaba };
\ No newline at end of file
+export { Alibaba };
diff --git a/app/components/chat.tsx b/app/components/chat.tsx
index 97e58da98..16a2a01b5 100644
--- a/app/components/chat.tsx
+++ b/app/components/chat.tsx
@@ -101,8 +101,6 @@ import {
 import { useNavigate } from "react-router-dom";
 import {
   CHAT_PAGE_SIZE,
-  DEFAULT_TTS_ENGINE,
-  ModelProvider,
   Path,
   REQUEST_TIMEOUT_MS,
   ServiceProvider,
@@ -1286,6 +1284,7 @@ function _Chat() {
   const accessStore = useAccessStore();
   const [speechStatus, setSpeechStatus] = useState(false);
   const [speechLoading, setSpeechLoading] = useState(false);
+  const [speechCooldown, setSpeechCooldown] = useState(false);
 
   async function openaiSpeech(text: string) {
     if (speechStatus) {
@@ -1297,10 +1296,10 @@ function _Chat() {
       api = new ClientApi(config.ttsConfig.modelProvider);
       setSpeechLoading(true);
       ttsPlayer.init();
-      let audioBuffer: ArrayBuffer;
+      let audioBuffer: ArrayBuffer | AudioBuffer;
       const { markdownToTxt } = require("markdown-to-txt");
       const textContent = markdownToTxt(text);
-      console.log("[OpenAI Speech] textContent: ", config, textContent);
+      console.log("[OpenAI Speech] textContent: ", textContent);
       if (config.ttsConfig.engine === "Edge") {
         const edgeVoiceName = accessStore.edgeVoiceName();
         const tts = new MsEdgeTTS();
@@ -1309,28 +1308,61 @@ function _Chat() {
           OUTPUT_FORMAT.AUDIO_24KHZ_96KBITRATE_MONO_MP3,
         );
         audioBuffer = await tts.toArrayBuffer(textContent);
+        playSpeech(audioBuffer);
       } else {
-        audioBuffer = await api.llm.speech({
-          model: config.ttsConfig.model,
-          input: textContent,
-          voice: config.ttsConfig.voice,
-          speed: config.ttsConfig.speed,
-        });
+        if (api.llm.streamSpeech) {
+          // 使用流式播放，边接收边播放
+          setSpeechStatus(true);
+          ttsPlayer.startStreamPlay(() => {
+            setSpeechStatus(false);
+          });
+
+          try {
+            for await (const chunk of api.llm.streamSpeech({
+              model: config.ttsConfig.model,
+              input: textContent,
+              voice: config.ttsConfig.voice,
+              speed: config.ttsConfig.speed,
+            })) {
+              console.log("[Stream Speech] add to queue", chunk);
+              ttsPlayer.addToQueue(chunk);
+            }
+            ttsPlayer.finishStreamPlay();
+          } catch (e) {
+            console.error("[Stream Speech]", e);
+            showToast(prettyObject(e));
+            setSpeechStatus(false);
+            ttsPlayer.stop();
+          } finally {
+            setSpeechLoading(false);
+          }
+        } else {
+          audioBuffer = await api.llm.speech({
+            model: config.ttsConfig.model,
+            input: textContent,
+            voice: config.ttsConfig.voice,
+            speed: config.ttsConfig.speed,
+          });
+          playSpeech(audioBuffer);
+        }
       }
-      setSpeechStatus(true);
-      ttsPlayer
-        .play(audioBuffer, () => {
-          setSpeechStatus(false);
-        })
-        .catch((e) => {
-          console.error("[OpenAI Speech]", e);
-          showToast(prettyObject(e));
-          setSpeechStatus(false);
-        })
-        .finally(() => setSpeechLoading(false));
     }
   }
 
+  function playSpeech(audioBuffer: ArrayBuffer | AudioBuffer) {
+    setSpeechStatus(true);
+    ttsPlayer
+      .play(audioBuffer, () => {
+        setSpeechStatus(false);
+      })
+      .catch((e) => {
+        console.error("[OpenAI Speech]", e);
+        showToast(prettyObject(e));
+        setSpeechStatus(false);
+      })
+      .finally(() => setSpeechLoading(false));
+  }
+
   const context: RenderMessage[] = useMemo(() => {
     return session.mask.hideContext ? [] : session.mask.context.slice();
   }, [session.mask.context, session.mask.hideContext]);
diff --git a/app/utils/audio.ts b/app/utils/audio.ts
index dd7eadc18..3d93f7bad 100644
--- a/app/utils/audio.ts
+++ b/app/utils/audio.ts
@@ -1,19 +1,38 @@
 type TTSPlayer = {
   init: () => void;
-  play: (audioBuffer: ArrayBuffer, onended: () => void | null) => Promise<void>;
+  play: (
+    audioBuffer: ArrayBuffer | AudioBuffer,
+    onended: () => void | null,
+  ) => Promise<void>;
+  playQueue: (
+    audioBuffers: (ArrayBuffer | AudioBuffer)[],
+    onended: () => void | null,
+  ) => Promise<void>;
+  addToQueue: (audioBuffer: ArrayBuffer | AudioBuffer) => void;
+  startStreamPlay: (onended: () => void | null) => void;
+  finishStreamPlay: () => void;
   stop: () => void;
 };
 
 export function createTTSPlayer(): TTSPlayer {
   let audioContext: AudioContext | null = null;
   let audioBufferSourceNode: AudioBufferSourceNode | null = null;
+  let isPlaying = false;
+  let playQueue: (ArrayBuffer | AudioBuffer)[] = [];
+  let currentOnended: (() => void | null) | null = null;
+  let isStreamMode = false;
+  let streamFinished = false;
 
   const init = () => {
+    console.log("[TTSPlayer] init");
     audioContext = new (window.AudioContext || window.webkitAudioContext)();
     audioContext.suspend();
   };
 
-  const play = async (audioBuffer: ArrayBuffer | AudioBuffer, onended: () => void | null) => {
+  const play = async (
+    audioBuffer: ArrayBuffer | AudioBuffer,
+    onended: () => void | null,
+  ) => {
     if (audioBufferSourceNode) {
       audioBufferSourceNode.stop();
       audioBufferSourceNode.disconnect();
@@ -33,17 +52,109 @@ export function createTTSPlayer(): TTSPlayer {
     audioBufferSourceNode.onended = onended;
   };
 
-  const stop = () => {
+  const playNext = async () => {
+    if (playQueue.length === 0) {
+      // 在流模式下，如果队列为空但流还没结束，等待
+      if (isStreamMode && !streamFinished) {
+        setTimeout(() => playNext(), 100);
+        return;
+      }
+
+      isPlaying = false;
+      isStreamMode = false;
+      streamFinished = false;
+      if (currentOnended) {
+        currentOnended();
+        currentOnended = null;
+      }
+      return;
+    }
+
+    const nextBuffer = playQueue.shift()!;
+    let buffer: AudioBuffer;
+    if (nextBuffer instanceof AudioBuffer) {
+      buffer = nextBuffer;
+    } else {
+      buffer = await audioContext!.decodeAudioData(nextBuffer);
+    }
+
+    if (audioBufferSourceNode) {
+      audioBufferSourceNode.stop();
+      audioBufferSourceNode.disconnect();
+    }
+
+    audioBufferSourceNode = audioContext!.createBufferSource();
+    audioBufferSourceNode.buffer = buffer;
+    audioBufferSourceNode.connect(audioContext!.destination);
+    audioBufferSourceNode.onended = () => {
+      playNext();
+    };
+
+    await audioContext!.resume();
+    audioBufferSourceNode.start();
+  };
+
+  const playQueueMethod = async (
+    audioBuffers: (ArrayBuffer | AudioBuffer)[],
+    onended: () => void | null,
+  ) => {
+    playQueue = [...audioBuffers];
+    currentOnended = onended;
+    if (!isPlaying) {
+      isPlaying = true;
+      await playNext();
+    }
+  };
+
+  const addToQueue = (audioBuffer: ArrayBuffer | AudioBuffer) => {
+    if (streamFinished) {
+      return;
+    }
+    playQueue.push(audioBuffer);
+  };
+
+  const startStreamPlay = (onended: () => void | null) => {
+    isStreamMode = true;
+    streamFinished = false;
+    playQueue = [];
+    currentOnended = onended;
+
+    if (!isPlaying) {
+      isPlaying = true;
+      playNext();
+    }
+  };
+
+  const finishStreamPlay = () => {
+    streamFinished = true;
+  };
+
+  const stop = async () => {
+    console.log("[TTSPlayer] stop");
+    playQueue = [];
+    isPlaying = false;
+    isStreamMode = false;
+    streamFinished = true;
+    currentOnended = null;
+
     if (audioBufferSourceNode) {
       audioBufferSourceNode.stop();
       audioBufferSourceNode.disconnect();
       audioBufferSourceNode = null;
     }
     if (audioContext) {
-      audioContext.close();
+      await audioContext.close();
       audioContext = null;
     }
   };
 
-  return { init, play, stop };
-}
\ No newline at end of file
+  return {
+    init,
+    play,
+    playQueue: playQueueMethod,
+    addToQueue,
+    startStreamPlay,
+    finishStreamPlay,
+    stop,
+  };
+}