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。
XML
<properties>
<project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
<javacv.version>1.5.6</javacv.version>
<system.windowsx64>windows-x86_64</system.windowsx64>
</properties>
<!-- javacv+javacpp -->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>javacv</artifactId>-->
<!-- <version>${javacv.version}</version>-->
<!-- </dependency>-->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>javacpp-platform</artifactId>-->
<!-- <version>${javacv.version}</version>-->
<!-- </dependency>-->
<!-- <!– ffmpeg最小依赖包，必须包含上面的javacv+javacpp核心库 –>-->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>ffmpeg</artifactId>-->
<!-- <version>4.4-${javacv.version}</version>-->
<!-- <classifier>${system.windowsx64}</classifier>-->
<!-- </dependency>-->
<!--<!– 最小opencv依赖包 ，必须包含上面的javacv+javacpp–>-->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>opencv</artifactId>-->
<!-- <version>4.5.1-${javacv.version}</version>-->
<!-- <classifier>${system.windowsx64}</classifier>-->
<!-- </dependency>-->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>openblas</artifactId>-->
<!-- <version>0.3.13-${javacv.version}</version>-->
<!-- <classifier>${system.windowsx64}</classifier>-->
<!-- </dependency>-->
<!-- <dependency>-->
<!-- <groupId>org.bytedeco</groupId>-->
<!-- <artifactId>flycapture</artifactId>-->
<!-- <version>2.13.3.31-${javacv.version}</version>-->
<!-- <classifier>${system.windowsx64}</classifier>-->
<!-- </dependency>-->
<dependencies>
<!-- 视频提取音频信息 -->
<dependency>
<groupId>org.bytedeco</groupId>
<artifactId>javacv-platform</artifactId>
<version>1.5.10</version>
</dependency>
<!-- 获取音频信息 -->
<dependency>
<groupId>org</groupId>
<artifactId>jaudiotagger</artifactId>
<version>2.0.3</version>
</dependency>
<dependency>
<groupId>net.java.dev.jna</groupId>
<artifactId>jna</artifactId>
<version>5.13.0</version>
</dependency>
<dependency>
<groupId>com.alphacephei</groupId>
<artifactId>vosk</artifactId>
<version>0.3.45</version>
</dependency>
<!-- JAVE2（Java音频视频编码器）库是ffmpeg项目上的Java包装器。 -->
<dependency>
<groupId>ws.schild</groupId>
<artifactId>jave-core</artifactId>
<version>3.1.1</version>
</dependency>
<dependency>
<groupId>com.alibaba</groupId>
<artifactId>fastjson</artifactId>
<version>1.2.83</version>
</dependency>
</dependencies>视频提取音频
java
package org.example;
import org.bytedeco.ffmpeg.global.avcodec;
import org.bytedeco.javacv.FFmpegFrameGrabber;
import org.bytedeco.javacv.FFmpegFrameRecorder;
import org.bytedeco.javacv.Frame;
public class Test {
public static void extractVoice(String sourceFileName, String audioUrl) throws FFmpegFrameGrabber.Exception, FFmpegFrameRecorder.Exception {
//抓取资源
FFmpegFrameGrabber frameGrabber = new FFmpegFrameGrabber(sourceFileName);
Frame frame = null;
FFmpegFrameRecorder recorder = null;
frameGrabber.start();
//转录为单轨, 16K采样率, wav格式
recorder = new FFmpegFrameRecorder(audioUrl, frameGrabber.getAudioChannels());
recorder.setFormat(frameGrabber.getFormat());
recorder.setSampleRate(frameGrabber.getSampleRate());//frameGrabber.getSampleRate()
//recorder.setAudioBitrate(128000);// 音频比特率
recorder.setTimestamp(frameGrabber.getTimestamp());
recorder.setVideoCodec(avcodec.AV_CODEC_ID_NONE); // 不录制视频
recorder.start();
int index = 0;
while (true) {
frame = frameGrabber.grabSamples();
if (frame == null) break;
if (frame.samples != null) {
recorder.recordSamples(frame.sampleRate, frame.audioChannels, frame.samples);
recorder.setTimestamp(frameGrabber.getTimestamp());
}
index++;
}
recorder.stop();
recorder.release();
frameGrabber.stop();
frameGrabber.release();
}
public static void main(String[] args) throws FFmpegFrameGrabber.Exception, FFmpegFrameRecorder.Exception {
String videoFilePath = "I:\\workspace\\test.mp4"; // 视频文件路径
String audioOutputPath = "I:\\workspace\\test_audio.wav"; // 输出的音频文件路径
long s = System.currentTimeMillis();
extractVoice(videoFilePath, audioOutputPath);
System.out.println(System.currentTimeMillis() - s);
}
}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
java
package org.example;
import com.alibaba.fastjson.JSON;
import org.vosk.LibVosk;
import org.vosk.LogLevel;
import org.vosk.Model;
import org.vosk.Recognizer;
import javax.sound.sampled.*;
import java.io.*;
import java.util.Optional;
public class Test3 {
public static void main(String[] args) {
StringBuilder result = new StringBuilder();
LibVosk.setLogLevel(LogLevel.DEBUG);
AudioFormat format = new AudioFormat(AudioFormat.Encoding.PCM_SIGNED, 44100, 16, 2, 4, 44100, false);
DataLine.Info info = new DataLine.Info(TargetDataLine.class, format);
TargetDataLine microphone;
SourceDataLine speakers;
try (Model model = new Model("I:\\workspace\\vosk-model-small-cn-0.22");
InputStream ais = AudioSystem.getAudioInputStream(new BufferedInputStream(new FileInputStream("I:\\workspace\\test_audio.wav")));
Recognizer recognizer = new Recognizer(model, 120000)) {
try {
microphone = (TargetDataLine) AudioSystem.getLine(info);
microphone.open(format);
microphone.start();
ByteArrayOutputStream out = new ByteArrayOutputStream();
int numBytesRead;
int CHUNK_SIZE = 1024;
int bytesRead = 0;
DataLine.Info dataLineInfo = new DataLine.Info(SourceDataLine.class, format);
speakers = (SourceDataLine) AudioSystem.getLine(dataLineInfo);
speakers.open(format);
speakers.start();
byte[] b = new byte[4096];
while (bytesRead <= 100000000) {
byte[] audioData = new byte[CHUNK_SIZE];
numBytesRead = ais.read(audioData, 0, CHUNK_SIZE);
bytesRead += numBytesRead;
out.write(audioData, 0, numBytesRead);
speakers.write(audioData, 0, numBytesRead);
if (recognizer.acceptWaveForm(audioData, numBytesRead)) {
result.append(getResult(recognizer.getResult()));
} else {
result.append(getResult(recognizer.getPartialResult()));
}
}
result.append(getResult(recognizer.getFinalResult()));
speakers.drain();
speakers.close();
microphone.close();
} catch (Exception e) {
e.printStackTrace();
}
System.out.println(result.toString());
} catch (IOException e) {
throw new RuntimeException(e);
} catch (UnsupportedAudioFileException e) {
throw new RuntimeException(e);
}
}
/**
* 获取返回结果
*
* @param result
* @return
*/
private static String getResult(String result) {
VoskResult vr = JSON.parseObject(result,VoskResult.class);
return Optional.ofNullable(vr).map(VoskResult::getText).orElse("");
}
public static void main1(String[] argv) throws IOException, UnsupportedAudioFileException {
LibVosk.setLogLevel(LogLevel.DEBUG);
StringBuilder result = new StringBuilder();
try (Model model = new Model("I:\\workspace\\vosk-model-small-cn-0.22");
InputStream ais = AudioSystem.getAudioInputStream(new BufferedInputStream(new FileInputStream("I:\\workspace\\test_audio.wav")));
Recognizer recognizer = new Recognizer(model, 120000)) {
int nbytes;
byte[] b = new byte[4096];
while ((nbytes = ais.read(b)) >= 0) {
if (recognizer.acceptWaveForm(b, nbytes)) {
result.append(getResult(recognizer.getResult()));
} else {
result.append(getResult(recognizer.getPartialResult()));
}
}
result.append(getResult(recognizer.getFinalResult()));
}
System.out.println(result);
}
}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。