devbox
人工智能uu
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

Java离线视频提取音频+音频提取文案_java解析音频中的文字

作者：人工智能uu | 2024-07-29 13:56:11

赞

踩

java解析音频中的文字

需引入依赖javacv、vosk相关依赖，

至于javacv依赖，网上有很多缩减方案，注释部分是可行的缩减方案，至于视频提取视频这里无需安装ffmpeg，只需引入依赖。而vosk需要下载模型方可使用，并且下载比较慢，可先用小模型跑通。

  1.     
  2.  
  3. <properties>
  4.     <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
  5.     <javacv.version>1.5.6</javacv.version>
  6.     <system.windowsx64>windows-x86_64</system.windowsx64>
  7.   </properties>
  8.  
  9. <!--  javacv+javacpp -->
  10. <!--    <dependency>-->
  11. <!--      <groupId>org.bytedeco</groupId>-->
  12. <!--      <artifactId>javacv</artifactId>-->
  13. <!--      <version>${javacv.version}</version>-->
  14. <!--    </dependency>-->
  15. <!--    <dependency>-->
  16. <!--      <groupId>org.bytedeco</groupId>-->
  17. <!--      <artifactId>javacpp-platform</artifactId>-->
  18. <!--      <version>${javacv.version}</version>-->
  19. <!--    </dependency>-->
  20. <!--    &lt;!&ndash; ffmpeg最小依赖包，必须包含上面的javacv+javacpp核心库 &ndash;&gt;-->
  21. <!--    <dependency>-->
  22. <!--      <groupId>org.bytedeco</groupId>-->
  23. <!--      <artifactId>ffmpeg</artifactId>-->
  24. <!--      <version>4.4-${javacv.version}</version>-->
  25. <!--      <classifier>${system.windowsx64}</classifier>-->
  26. <!--    </dependency>-->
  27. <!--&lt;!&ndash;     最小opencv依赖包 ，必须包含上面的javacv+javacpp&ndash;&gt;-->
  28. <!--        <dependency>-->
  29. <!--          <groupId>org.bytedeco</groupId>-->
  30. <!--          <artifactId>opencv</artifactId>-->
  31. <!--          <version>4.5.1-${javacv.version}</version>-->
  32. <!--          <classifier>${system.windowsx64}</classifier>-->
  33. <!--        </dependency>-->
  34. <!--        <dependency>-->
  35. <!--          <groupId>org.bytedeco</groupId>-->
  36. <!--          <artifactId>openblas</artifactId>-->
  37. <!--          <version>0.3.13-${javacv.version}</version>-->
  38. <!--          <classifier>${system.windowsx64}</classifier>-->
  39. <!--        </dependency>-->
  40. <!--    <dependency>-->
  41. <!--      <groupId>org.bytedeco</groupId>-->
  42. <!--      <artifactId>flycapture</artifactId>-->
  43. <!--      <version>2.13.3.31-${javacv.version}</version>-->
  44. <!--      <classifier>${system.windowsx64}</classifier>-->
  45. <!--    </dependency>-->
  46.  
  47.   <dependencies>
  48.     <!-- 视频提取音频信息 -->
  49.     <dependency>
  50.       <groupId>org.bytedeco</groupId>
  51.       <artifactId>javacv-platform</artifactId>
  52.       <version>1.5.10</version>
  53.     </dependency>
  54.  
  55.     <!-- 获取音频信息 -->
  56.     <dependency>
  57.       <groupId>org</groupId>
  58.       <artifactId>jaudiotagger</artifactId>
  59.       <version>2.0.3</version>
  60.     </dependency>
  61.     <dependency>
  62.       <groupId>net.java.dev.jna</groupId>
  63.       <artifactId>jna</artifactId>
  64.       <version>5.13.0</version>
  65.     </dependency>
  66.     <dependency>
  67.       <groupId>com.alphacephei</groupId>
  68.       <artifactId>vosk</artifactId>
  69.       <version>0.3.45</version>
  70.     </dependency>
  71.     <!-- JAVE2（Java音频视频编码器）库是ffmpeg项目上的Java包装器。 -->
  72.     <dependency>
  73.       <groupId>ws.schild</groupId>
  74.       <artifactId>jave-core</artifactId>
  75.       <version>3.1.1</version>
  76.     </dependency>
  77.  
  78.     <dependency>
  79.       <groupId>com.alibaba</groupId>
  80.       <artifactId>fastjson</artifactId>
  81.       <version>1.2.83</version>
  82.     </dependency>
  83.   </dependencies>

视频提取音频

  1. package org.example;
  2.  
  3. import org.bytedeco.ffmpeg.global.avcodec;
  4. import org.bytedeco.javacv.FFmpegFrameGrabber;
  5. import org.bytedeco.javacv.FFmpegFrameRecorder;
  6. import org.bytedeco.javacv.Frame;
  7.  
  8.  
  9. public class Test {
  10.  
  11.     public static void extractVoice(String sourceFileName, String audioUrl) throws FFmpegFrameGrabber.Exception, FFmpegFrameRecorder.Exception {
  12.         //抓取资源
  13.         FFmpegFrameGrabber frameGrabber = new FFmpegFrameGrabber(sourceFileName);
  14.         Frame frame = null;
  15.         FFmpegFrameRecorder recorder = null;
  16.         frameGrabber.start();
  17.         //转录为单轨, 16K采样率, wav格式
  18.         recorder = new FFmpegFrameRecorder(audioUrl, frameGrabber.getAudioChannels());
  19.         recorder.setFormat(frameGrabber.getFormat());
  20.         recorder.setSampleRate(frameGrabber.getSampleRate());//frameGrabber.getSampleRate()
  21.         //recorder.setAudioBitrate(128000);// 音频比特率
  22.         recorder.setTimestamp(frameGrabber.getTimestamp());
  23.         recorder.setVideoCodec(avcodec.AV_CODEC_ID_NONE); // 不录制视频
  24.  
  25.         recorder.start();
  26.         int index = 0;
  27.         while (true) {
  28.             frame = frameGrabber.grabSamples();
  29.             if (frame == null) break;
  30.             if (frame.samples != null) {
  31.                 recorder.recordSamples(frame.sampleRate, frame.audioChannels, frame.samples);
  32.                 recorder.setTimestamp(frameGrabber.getTimestamp());
  33.             }
  34.             index++;
  35.         }
  36.         recorder.stop();
  37.         recorder.release();
  38.         frameGrabber.stop();
  39.         frameGrabber.release();
  40.     }
  41.  
  42.     public static void main(String[] args) throws FFmpegFrameGrabber.Exception, FFmpegFrameRecorder.Exception {
  43.         String videoFilePath = "I:\\workspace\\test.mp4"; // 视频文件路径
  44.         String audioOutputPath = "I:\\workspace\\test_audio.wav"; // 输出的音频文件路径
  45.         long s = System.currentTimeMillis();
  46.         extractVoice(videoFilePath, audioOutputPath);
  47.         System.out.println(System.currentTimeMillis() - s);
  48.     }
  49.  
  50. }

音频提取文字

至于model可去此网站下载，解压使用。大模型下载较慢

VOSK Models

  1. package org.example;
  2.  
  3. import com.alibaba.fastjson.JSON;
  4. import org.vosk.LibVosk;
  5. import org.vosk.LogLevel;
  6. import org.vosk.Model;
  7. import org.vosk.Recognizer;
  8.  
  9. import javax.sound.sampled.*;
  10. import java.io.*;
  11. import java.util.Optional;
  12.  
  13. public class Test3 {
  14.     public static void main(String[] args) {
  15.         StringBuilder result = new StringBuilder();
  16.         LibVosk.setLogLevel(LogLevel.DEBUG);
  17.  
  18.             AudioFormat format = new AudioFormat(AudioFormat.Encoding.PCM_SIGNED, 44100, 16, 2, 4, 44100, false);
  19.         DataLine.Info info = new DataLine.Info(TargetDataLine.class, format);
  20.         TargetDataLine microphone;
  21.         SourceDataLine speakers;
  22.  
  23.         try (Model model = new Model("I:\\workspace\\vosk-model-small-cn-0.22");
  24.              InputStream ais = AudioSystem.getAudioInputStream(new BufferedInputStream(new FileInputStream("I:\\workspace\\test_audio.wav")));
  25.              Recognizer recognizer = new Recognizer(model, 120000)) {
  26.             try {
  27.  
  28.                 microphone = (TargetDataLine) AudioSystem.getLine(info);
  29.                 microphone.open(format);
  30.                 microphone.start();
  31.  
  32.                 ByteArrayOutputStream out = new ByteArrayOutputStream();
  33.                 int numBytesRead;
  34.                 int CHUNK_SIZE = 1024;
  35.                 int bytesRead = 0;
  36.  
  37.                 DataLine.Info dataLineInfo = new DataLine.Info(SourceDataLine.class, format);
  38.                 speakers = (SourceDataLine) AudioSystem.getLine(dataLineInfo);
  39.                 speakers.open(format);
  40.                 speakers.start();
  41.                 byte[] b = new byte[4096];
  42.  
  43.                 while (bytesRead <= 100000000) {
  44.                     byte[] audioData = new byte[CHUNK_SIZE];
  45.                     numBytesRead = ais.read(audioData, 0, CHUNK_SIZE);
  46.                     bytesRead += numBytesRead;
  47.  
  48.                     out.write(audioData, 0, numBytesRead);
  49.  
  50.                     speakers.write(audioData, 0, numBytesRead);
  51.  
  52.                     if (recognizer.acceptWaveForm(audioData, numBytesRead)) {
  53.                         result.append(getResult(recognizer.getResult()));
  54.                     } else {
  55.                         result.append(getResult(recognizer.getPartialResult()));
  56.                     }
  57.                 }
  58.                 result.append(getResult(recognizer.getFinalResult()));
  59.                 speakers.drain();
  60.                 speakers.close();
  61.                 microphone.close();
  62.             } catch (Exception e) {
  63.                 e.printStackTrace();
  64.             }
  65.             System.out.println(result.toString());
  66.         } catch (IOException e) {
  67.             throw new RuntimeException(e);
  68.         } catch (UnsupportedAudioFileException e) {
  69.             throw new RuntimeException(e);
  70.         }
  71.     }
  72.  
  73.     /**
  74.      * 获取返回结果
  75.      *
  76.      * @param result
  77.      * @return
  78.      */
  79.     private static String getResult(String result) {
  80.         VoskResult vr = JSON.parseObject(result,VoskResult.class);
  81.         return  Optional.ofNullable(vr).map(VoskResult::getText).orElse("");
  82.  
  83.     }
  84.  
  85.     public static void main1(String[] argv) throws IOException, UnsupportedAudioFileException {
  86.         LibVosk.setLogLevel(LogLevel.DEBUG);
  87.         StringBuilder result = new StringBuilder();
  88.         try (Model model = new Model("I:\\workspace\\vosk-model-small-cn-0.22");
  89.              InputStream ais = AudioSystem.getAudioInputStream(new BufferedInputStream(new FileInputStream("I:\\workspace\\test_audio.wav")));
  90.              Recognizer recognizer = new Recognizer(model, 120000)) {
  91.  
  92.             int nbytes;
  93.             byte[] b = new byte[4096];
  94.             while ((nbytes = ais.read(b)) >= 0) {
  95.                 if (recognizer.acceptWaveForm(b, nbytes)) {
  96.                     result.append(getResult(recognizer.getResult()));
  97.                 } else {
  98.                     result.append(getResult(recognizer.getPartialResult()));
  99.                 }
  100.             }
  101.             result.append(getResult(recognizer.getFinalResult()));
  102.         }
  103.         System.out.println(result);
  104.     }
  105. }

感谢网上各位大佬能分享这些信息

测试可行，识别率没有做过对比、大模型也没有试过。这里也就提供一种可行的离线解决方案。

声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/人工智能uu/article/detail/899042
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号