语音识别服务funasr搭建

以微薄之力推动正能量 發表於 2025-12-26 09:31:00

本文讨论语音识别功能，使用的是阿里的开源语音识别项目FunASR，含两种部署方式，社区windows版和docker容器化部署，windows社区版的可以用于本地开发使用，生产环境建议使用容器版。
<h2>1、windows社区版部署</h2>
<h3>　　1.1、环境安装</h3>
　　　　软件需要Visual Studio 2022 c++环境，如果没有Visual Studio 2022 c++运行环境，双击 VC_redist.x64(2022).exe 安装 Visual Studio 2022环境下编译的C++程序运行所需要的库。
<h3>　　 1.2、下载windows社区软件包 </h3>
　　　　https://www.modelscope.cn/models/iic/funasr-runtime-win-cpu-x64/files
　　　　<img alt="image" width="1253" height="478" loading="lazy" src="https://img2024.cnblogs.com/blog/1607557/202512/1607557-20251226091034169-345184089.png" class="lazyload">
　　　　随便选个版本的下载，这里选择的是0.2.0版本
<h3>　　 1.3、下载所需模型</h3>
<div class="cnblogs_code">
<pre>git clone https://www.modelscope.cn/damo/speech_fsmn_vad_zh-cn-16k-common-onnx.git;

git clone https://www.modelscope.cn/damo/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-onnx.git;

git clone https://www.modelscope.cn/damo/speech_ngram_lm_zh-cn-ai-wesp-fst.git;

git clone https://www.modelscope.cn/damo/punc_ct-transformer_cn-en-common-vocab471067-large-onnx.git;

git clone https://www.modelscope.cn/thuduj12/fst_itn_zh.git</pre>
</div>
<h3>　　 1.4、启动服务</h3>
 　　　　将上面下载的windows社区软件包解压后，打开powershell，进入到解压后的目录，执行下面的命令
<div class="cnblogs_code">
<pre>./funasr-wss-server.exe
--vad-dir D:/developTest/funasr-runtime-resources/models/speech_fsmn_vad_zh-cn-16k-common-onnx
--model-dir D:/developTest/funasr-runtime-resources/models/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-onnx
--lm-dir D:/developTest/funasr-runtime-resources/models/speech_ngram_lm_zh-cn-ai-wesp-fst
--punc-dir D:/developTest/funasr-runtime-resources/models/punc_ct-transformer_cn-en-common-vocab471067-large-onnx
--itn-dir D:/developTest/funasr-runtime-resources/models/fst_itn_zh
--certfile 0</pre>
</div>
　　　　参数说明：
<div class="cnblogs_code">
<pre>--model-dirmodelscope model ID 或者本地模型路径
--vad-dirmodelscope model ID 或者本地模型路径
--punc-dirmodelscope model ID 或者本地模型路径
--lm-dir modelscope model ID 或者本地模型路径
--itn-dir modelscope model ID 或者本地模型路径
--certfilessl的证书文件，如果需要关闭ssl，参数设置为0</pre>
</div>
<h3>　　 1.5、客户端调用</h3>
　　　　在windows社区版的解压目录下有客户端执行文件funasr-wss-client.exe
<div class="cnblogs_code">
<pre>./funasr-wss-client.exe --server-ip 127.0.0.1 --port 10095 --wav-path asr_example_zh.wav</pre>
</div>
　　　　服务默认端口是10095，--wav-path指定音频文件地址
<h2>2、docker容器化部署</h2>
<h3>　　 2.1、拉取docker镜像</h3>
<div class="cnblogs_code">
<pre>docker pull registry.cn-hangzhou.aliyuncs.com/funasr_repo/funasr:funasr-runtime-sdk-cpu-0.4.6</pre>
</div>
<h3>　　 2.2、启动容器</h3>
　　　　在宿主机创建模型目录放置模型，这里的模型建议手动下载，就用上面的git下载下来，如果使用启动命令自动下载会很慢很卡。
<div class="cnblogs_code">
<pre>docker run -p 10095:10095 -it --privileged=true -v D:\developTest\funasr-runtime-resources\models:/workspace/models registry.cn-hangzhou.aliyuncs.com/funasr_repo/funasr:funasr-runtime-sdk-cpu-0.4.6</pre>
</div>
　　　　映射容器端口，挂载之前创建的存放模型目录到容器内部。
<h3>　　 2.3、启动服务</h3>
　　　　进入容器内部，进到FunASR/runtime目录下
　　　　<img alt="image" loading="lazy" src="https://img2024.cnblogs.com/blog/1607557/202512/1607557-20251226092232599-408304639.png" class="lazyload">
 　　　　执行如下命令启动服务
<div class="cnblogs_code">
<pre>nohup bash run_server.sh \
--certfile 0 \
--vad-dir speech_fsmn_vad_zh-cn-16k-common-onnx \
--model-dir speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-onnx \
--punc-dir punc_ct-transformer_cn-en-common-vocab471067-large-onnx\
--lm-dir speech_ngram_lm_zh-cn-ai-wesp-fst \
--itn-dir fst_itn_zh> log.txt 2>&1 &</pre>
</div>
　　　　指定模型目录，这里的模型都是事先下载好的，就不需要通过启动命令下载了，certfile设为0，表示关闭ssl。
<h2>3、调用示例</h2>
　　这里大致写了两种java调用方式一种是通过ProcessBuilder，一种是WebSocketClient，大家可以用来看看。
<div class="lake-content">
<ul class="ne-ul">
<li id="u0a0bed12" data-lake-index-type="0">使用ProcessBuilder，运行上面的客户端执行命令，获取执行结果</li>
</ul>
<div class="cnblogs_code"><img id="code_img_closed_ee8da724-9620-48b4-9e11-58304394a99c" class="code_img_closed lazyload" data-src="http://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif"><img id="code_img_opened_ee8da724-9620-48b4-9e11-58304394a99c" class="code_img_opened lazyload" style="display: none" data-src="http://images.cnblogs.com/OutliningIndicators/ExpandedBlockStart.gif">
<div id="cnblogs_code_open_ee8da724-9620-48b4-9e11-58304394a99c" class="cnblogs_code_hide">
<pre>public String localTranslation(MultipartRequest multipartRequest) {
 StringBuffer resultBuffer = new StringBuffer();
 // 需要传递给exe程序的参数
 String exePath = "D:\\developTest\\funasr-runtime-resources\\funasr-runtime-win-cpu-x64\\funasr-runtime-win-cpu-x64-v0.2.0\\funasr-wss-client.exe";
 String serveIp = "127.0.0.1"; // 假设你想要设置的IP地址
 String port = "10095";
 File targetFile = null;

 try {
 MultipartFile mFile = multipartRequest.getFile("file");
 File dir = new File("D:\\developTest\\funasr-runtime-resources\\wav");
 if (!dir.exists()) {
 dir.mkdirs();
 }
 targetFile = File.createTempFile("tmp_", ".wav", dir);
 mFile.transferTo(targetFile);

 String wavPath = "D:\\developTest\\funasr-runtime-resources\\wav\\"+ targetFile.getName();
 String[] cmd = new String[]{exePath, "--server-ip", serveIp, "--port", port, "--wav-path", wavPath};
 ProcessBuilder pb = new ProcessBuilder();
 pb.command(cmd);

 Process process = pb.start();
 //超时时间
 int timeoutSeconds = 30;//超时30秒自动断开

 //创建单线程线程池
 ExecutorService executor = Executors.newSingleThreadExecutor();
 Future<?> future = executor.submit(() -> {
 try {
 pb.redirectErrorStream(true);
 // 读取外部程序的输出
 BufferedReader reader = new BufferedReader(new InputStreamReader(process.getInputStream()));
 String line;
 while ((line = reader.readLine()) != null) {
 System.out.println(line);
 resultBuffer.append(line);
 }

 // 处理错误输出
 BufferedReader errorReader = new BufferedReader(new InputStreamReader(process.getErrorStream()));
 while ((line = errorReader.readLine()) != null) {
 System.out.println(line);
 if(line.contains("on_message")){
 String[] array = line.split("on_message =");
 resultBuffer.append(array[1]);
 }
 }
 // 等待程序执行完成
 process.waitFor();
 } catch (Exception e) {
 e.printStackTrace();
 }finally {
 if(process.isAlive()){
 process.destroy();
 }
 }
 });
 try {
 // 等待进程完成或超时
 future.get(timeoutSeconds, TimeUnit.SECONDS);
 System.out.println("进程在规定时间内完成。");
 } catch (Exception e) {
 System.out.println("超时预警: 进程可能挂起。");
 resultBuffer.append("timeout");
 } finally {
 //关闭连接
 if(process.isAlive()){
 process.destroy();
 }
 executor.shutdownNow(); // 取消任务并关闭线程池
 }
 } catch (Exception e) {
 e.printStackTrace();
 resultBuffer.append("error");
 }finally {
 if (targetFile.exists()) {
 targetFile.delete();
 }
 }
 System.out.println(resultBuffer.toString());
 return resultBuffer.toString();
}</pre>
</div>
View Code</div>
<div class="lake-content">
<ul class="ne-ul">
<li id="u199a6f15" data-lake-index-type="0">使用WebSocketClient直接调用FunASR服务</li>
</ul>
　　Client工具类
<div class="cnblogs_code"><img id="code_img_closed_b0c7855c-8229-4a70-a43a-7df98e723287" class="code_img_closed lazyload" data-src="http://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif"><img id="code_img_opened_b0c7855c-8229-4a70-a43a-7df98e723287" class="code_img_opened lazyload" style="display: none" data-src="http://images.cnblogs.com/OutliningIndicators/ExpandedBlockStart.gif">
<div id="cnblogs_code_open_b0c7855c-8229-4a70-a43a-7df98e723287" class="cnblogs_code_hide">
<pre>package com.example.demo1.web;

import java.io.*;
import java.net.URI;
import java.util.Map;
import org.java_websocket.client.WebSocketClient;
import org.java_websocket.drafts.Draft;
import org.java_websocket.handshake.ServerHandshake;
import org.json.simple.JSONArray;
import org.json.simple.JSONObject;
import org.json.simple.parser.JSONParser;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.regex.Matcher;
import java.util.regex.Pattern;

public class FunasrWsClient extends WebSocketClient {

private static final Logger logger = LoggerFactory.getLogger(FunasrWsClient.class);

private boolean iseof = false;
private static String wavPath;
private static String mode = "offline";
private static String strChunkSize = "5,10,5";
private static int chunkInterval = 10;
private static int sendChunkSize = 1920;
private static String hotwords="";
private static String fsthotwords="";
private String wavName = "javatest";
private MyCallBack callBack;

public FunasrWsClient(URI serverUri,MyCallBack callBack) {
 super(serverUri);
 this.callBack = callBack;
}

public FunasrWsClient(URI serverUri,String wavPath,MyCallBack callBack) {
 super(serverUri);
 this.callBack = callBack;
 this.wavPath = wavPath;
}

public FunasrWsClient(URI serverUri,String strChunkSize,int chunkInterval,String mode,String hotwords,String wavPath,MyCallBack callBack) {
 super(serverUri);
 this.callBack = callBack;
 this.strChunkSize = strChunkSize;
 this.chunkInterval = chunkInterval;
 this.mode = mode;
 this.fsthotwords = hotwords;
 this.wavPath = wavPath;

 int RATE = 16000;
 String[] chunkList = strChunkSize.split(",");
 int int_chunk_size = 60 * Integer.valueOf(chunkList.trim()) / chunkInterval;
 int CHUNK = Integer.valueOf(RATE / 1000 * int_chunk_size);
 this.sendChunkSize = CHUNK * 2;
}

public class RecWavThread extends Thread {
 private FunasrWsClient funasrClient;

 public RecWavThread(FunasrWsClient funasrClient) {
 this.funasrClient = funasrClient;
 }

 public void run() {
 this.funasrClient.recWav();
 }
}

public FunasrWsClient(URI serverUri, Draft draft) {
 super(serverUri, draft);
}

public FunasrWsClient(URI serverURI) {

 super(serverURI);
}

public FunasrWsClient(URI serverUri, Map<String, String> httpHeaders) {
 super(serverUri, httpHeaders);
}

public void getSslContext(String keyfile, String certfile) {
 // TODO
 return;
}

public void sendJson(
 String mode, String strChunkSize, int chunkInterval, String wavName, boolean isSpeaking,String suffix) {
 try {

 JSONObject obj = new JSONObject();
 obj.put("mode", mode);
 JSONArray array = new JSONArray();
 String[] chunkList = strChunkSize.split(",");
 for (int i = 0; i < chunkList.length; i++) {
 array.add(Integer.valueOf(chunkList.trim()));
 }

 obj.put("chunk_size", array);
 obj.put("chunk_interval", new Integer(chunkInterval));
 obj.put("wav_name", wavName);

 if(FunasrWsClient.hotwords.trim().length()>0)
 {
 String regex = "\\d+";
 JSONObject jsonitems = new JSONObject();
 String[] items=FunasrWsClient.hotwords.trim().split(" ");
 Pattern pattern = Pattern.compile(regex);
 String tmpWords="";
 for(int i=0;i<items.length;i++)
 {

 Matcher matcher = pattern.matcher(items);

 if (matcher.matches()) {

 jsonitems.put(tmpWords.trim(), items.trim());
 tmpWords="";
 continue;
 }
 tmpWords=tmpWords+items+" ";

 }

 obj.put("hotwords", jsonitems.toString());
 }

 if(suffix.equals("wav")){
 suffix="pcm";
 }
 obj.put("wav_format", suffix);
 if (isSpeaking) {
 obj.put("is_speaking", new Boolean(true));
 } else {
 obj.put("is_speaking", new Boolean(false));
 }
 logger.info("sendJson: " + obj);
 // return;

 send(obj.toString());

 return;
 } catch (Exception e) {
 e.printStackTrace();
 }
}

public void sendEof() {
 try {
 JSONObject obj = new JSONObject();

 obj.put("is_speaking", new Boolean(false));

 logger.info("sendEof: " + obj);
 // return;

 send(obj.toString());
 iseof = true;
 return;
 } catch (Exception e) {
 e.printStackTrace();
 }
}

public void recWav() {
 String fileName=FunasrWsClient.wavPath;
 String suffix=fileName.split("\\.");
 sendJson(mode, strChunkSize, chunkInterval, wavName, true,suffix);
 File file = new File(FunasrWsClient.wavPath);

 int chunkSize = sendChunkSize;
 byte[] bytes = new byte;

 int readSize = 0;
 try (FileInputStream fis = new FileInputStream(file)) {
 if (FunasrWsClient.wavPath.endsWith(".wav")) {
 fis.read(bytes, 0, 44);
 }
 readSize = fis.read(bytes, 0, chunkSize);
 while (readSize > 0) {
 if (readSize == chunkSize) {
 send(bytes);

 } else {
 byte[] tmpBytes = new byte;
 for (int i = 0; i < readSize; i++) {
 tmpBytes = bytes;
 }
 send(tmpBytes);
 }

 if (!mode.equals("offline")) {
 Thread.sleep(Integer.valueOf(chunkSize / 32));
 }

 readSize = fis.read(bytes, 0, chunkSize);
 }

 if (!mode.equals("offline")) {
 Thread.sleep(2000);
 sendEof();
 Thread.sleep(3000);
 close();
 } else {
 sendEof();
 }

 } catch (Exception e) {
 e.printStackTrace();
 }
}

@Override
public void onOpen(ServerHandshake handshakedata) {

 RecWavThread thread = new RecWavThread(this);
 thread.start();
}

@Override
public void onMessage(String message) {
 JSONObject jsonObject = new JSONObject();
 JSONParser jsonParser = new JSONParser();
 logger.info("received: " + message);
 try {
 jsonObject = (JSONObject) jsonParser.parse(message);
 logger.info("text: " + jsonObject.get("text"));
 callBack.callBack(jsonObject.get("text"));
 if(jsonObject.containsKey("timestamp"))
 {
 logger.info("timestamp: " + jsonObject.get("timestamp"));
 }
 } catch (org.json.simple.parser.ParseException e) {
 e.printStackTrace();
 }
 if (iseof && mode.equals("offline") && !jsonObject.containsKey("is_final")) {
 close();
 }

 if (iseof && mode.equals("offline") && jsonObject.containsKey("is_final") && jsonObject.get("is_final").equals("false")) {
 close();
 }
}

@Override
public void onClose(int code, String reason, boolean remote) {

 logger.info(
 "Connection closed by "
 + (remote ? "remote peer" : "us")
 + " Code: "
 + code
 + " Reason: "
 + reason);
}

@Override
public void onError(Exception ex) {
 logger.info("ex: " + ex);
 ex.printStackTrace();
}

}</pre>
</div>
View Code</div>
<div class="lake-content">　　调用工具类</div>
<div class="lake-content">
<div class="cnblogs_code"><img id="code_img_closed_2af70ca2-29d6-46e7-92ec-d8fd7d13b2df" class="code_img_closed lazyload" data-src="http://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif"><img id="code_img_opened_2af70ca2-29d6-46e7-92ec-d8fd7d13b2df" class="code_img_opened lazyload" style="display: none" data-src="http://images.cnblogs.com/OutliningIndicators/ExpandedBlockStart.gif">
<div id="cnblogs_code_open_2af70ca2-29d6-46e7-92ec-d8fd7d13b2df" class="cnblogs_code_hide">
<pre>public static void main(String[] args) throws URISyntaxException {
String srvIp = "localhost";
String srvPort = "10095";
String wavPath = "D:\\developTest\\funasr-runtime-resources\\wav\\tmp_84677349854990998.wav";
Object lock = new Object();

StringBuffer text = new StringBuffer();
ExecutorService executor = Executors.newSingleThreadExecutor();
Future<?> future = executor.submit(()->{
 try {
 String wsAddress = "ws://" + srvIp + ":" + srvPort;

 FunasrWsClient c = new FunasrWsClient(new URI(wsAddress),wavPath,new MyCallBack(){
 @Override
 public void callBack(Object obj){
 text.append(obj.toString());
 synchronized (lock){
 try {
 lock.notify();
 }catch (Exception e){
 e.printStackTrace();
 }
 }
 }
 });

 synchronized (lock){
 c.connect();
 lock.wait();
 }
 }catch (Exception e){
 // e.printStackTrace();
 }
});
try {
 future.get(10, TimeUnit.SECONDS);
 System.out.println("规定时间内完成");
}catch (Exception e){
 // e.printStackTrace();
 System.out.println("任务超时");
 text.append("任务超时");
}finally {
 executor.shutdownNow(); // 取消任务并关闭线程池
 }
System.out.println(text.toString());
}</pre>
</div>
View Code</div>
 
</div>
<div class="lake-content">
 
</div>
</div>
</div>
  
来源：https://www.cnblogs.com/zhouyun-yx/p/19402607

頁: [1]

琼殿技术论坛's Archiver

语音识别服务funasr搭建