Unity语音唤醒与识别

在Unity中实现语音唤醒与识别功能

先描述一下该模块的实现方式：通过监听麦克风音量获取实时音量，设置一个音量阈值，当音量超过阈值后开始记录从麦克风获取的音频数据，在获取的数据前加上PCM文件头，然后直接上传给阿里识别平台，获取识别结果。

1. 初始化麦克风

   private void InitMicphone()
   {
       if (Microphone.devices.Length > 0)
       {
           curDevice = Microphone.devices[0];
           Debug.Log("当前Mic设备 : " + curDevice);
   
           if (Microphone.devices[0].IsNormalized())
           {
               micRecords = Microphone.Start(curDevice, true, 60, 16000);
               Debug.Log("Mic start!");
           }
       }
       else
       {
           Debug.Log("Mic missing!");
           return;
       }
   }

2. 获取实时音量

   private float GetMaxVolume()
   {
       float maxVolume = 0f;
   
       float[] volumeData = new float[128];
       int offset = Microphone.GetPosition(null) - 128 + 1;
       if (offset < 0)
       {
           return 0f;
       }
       micRecords.GetData(volumeData, offset);
   
       for (int i = 0; i < 128; i++)
       {
           float tempMax = volumeData[i];
           if (maxVolume < tempMax)
           {
               maxVolume = tempMax;
           }
       }
   
       return maxVolume;
   }

   volumeData数组的长度取决于你想要识别的频率，越小识别频率越高，当然性能损耗也越高。相当于吧这个数组内的数据看做“一帧”，识别这一帧的最大音量作为实时音量返回。
3. 写PCM文件头

   private void WriteHeader(AudioClip clip)
   {
       int hz = clip.frequency;
       int channels = clip.channels;
       int samples = clip.samples;
   
       bytesList.Clear();
   
       Byte[] riff = System.Text.Encoding.UTF8.GetBytes("RIFF");
       bytesList.AddRange(riff);
   
       Byte[] chunkSize = BitConverter.GetBytes(36);   // 44 - 8
       bytesList.AddRange(chunkSize);
   
       Byte[] wave = System.Text.Encoding.UTF8.GetBytes("WAVE");
       bytesList.AddRange(wave);
   
       Byte[] fmt = System.Text.Encoding.UTF8.GetBytes("fmt ");
       bytesList.AddRange(fmt);
   
       Byte[] subChunk = BitConverter.GetBytes(16);
       bytesList.AddRange(subChunk);
   
       Byte[] audioFormat = BitConverter.GetBytes(1);
       bytesList.AddRange(audioFormat);
   
       Byte[] channel = BitConverter.GetBytes(channels);
       bytesList.AddRange(channel);
   
       Byte[] sampleRate = BitConverter.GetBytes(hz);
       bytesList.AddRange(sampleRate);
   
       Byte[] byteRate = BitConverter.GetBytes(hz * channels * 2);
       bytesList.AddRange(byteRate);
   
       Byte[] blockAlign = BitConverter.GetBytes(channels * 2);
       bytesList.AddRange(blockAlign);
   
       Byte[] bitsPerSample = BitConverter.GetBytes(16);
       bytesList.AddRange(bitsPerSample);
   
       Byte[] data = System.Text.Encoding.UTF8.GetBytes("data");
       bytesList.AddRange(data);
   
       Byte[] subChunk2 = BitConverter.GetBytes(samples * channels * 2);
       bytesList.AddRange(subChunk2);
   }

   这里写进去的内容都是上一篇文章中提到的PCM文件头相关的信息，具体可以参见上一篇PCM文件头解析。
4. 写音频数据体

   private void WriteBody(AudioClip clip)
   {
       float[] samples = new float[clip.samples];
   
       clip.GetData(samples, 0);
   
       Int16[] intData = new Int16[samples.Length];
   
       Byte[] bytesData = new Byte[samples.Length * 2];
   
       int rescaleFactor = 32767; //to convert float to Int16 
   
       for (int i = 0; i < samples.Length; i++)
       {
           intData[i] = (short)(samples[i] * rescaleFactor);
           Byte[] byteArr = new Byte[2];
           byteArr = BitConverter.GetBytes(intData[i]);
           byteArr.CopyTo(bytesData, i * 2);
       }
   
       bytesList.AddRange(bytesData);
   }

   传进来的 AudioClip 为达到阈值后录下来的音频数据，将其转换为二进制数据与文件头拼接，就可以得到完整的音频数据。
5. 获取音频数据

   private void GetSpeechData()
   {
       if (Microphone.GetPosition(null) - curPosition < responseTime)
       {
           return;
       }
       float[] volumeData = new float[responseTime];
       micRecords.GetData(volumeData, (int)curPosition);
   
       WriteHeader(micRecords);
       WriteBody(micRecords);
   
       SpeechRecognizer.GetSpeechResult("pcm", 16000, bytesList.ToArray());
       isRecording = false;
   }

   获取数据并包装成完整音频数据然后上传识别平台。
   SpeechRecognizer.GetSpeechResult 函数就是阿里音频识别相关的内容了，因为接口会变，所以具体如何写参见阿里对应的文档吧。