基于C#的在线语音识别软件开发

本软件利用了百度语音识别提供的接口，自行开发出的一个在线的语音识别软件。所以，制作之前需要去百度语音识别的网站去注册一个项目，免费的除非你的需求量特别大不然不需要付费。百度语音识别地址

然后就需要自己写代码去解决以下问题

获取麦克风输入的语音

发送到百度语音识别的接口

得到返还的信息识别。

获取麦克风输入的语音

要获取麦克风的输入，需要调用一些WindowsAPI及其他的东西。下面就慢慢梳理 我会分散的梳理，整合需要自己理解着去整合

首先，我们获取麦克风，使用winmm.dll

//调用wavein的dll
[DllImport("winmm.dll")]
//获取有多少可用输入设备
public static extern int waveInGetNumDevs();
[DllImport("winmm.dll")]
//增加一个缓冲区
public static extern int waveInAddBuffer(IntPtr hwi, ref WaveHdr pwh, UInt32 cbwh);
[DllImport("winmm.dll")]
//关闭麦克风
public static extern int waveInClose(IntPtr hwi);
[DllImport("winmm.dll")]
//打开麦克风
public static extern int waveInOpen(out IntPtr phwi, UInt32 uDeviceID, ref WaveFormatEx lpFormat, WaveDelegate dwCallback, UInt32 dwInstance, UInt32 dwFlags);
[DllImport("winmm.dll")]
//标记为可用的缓冲区
public static extern int waveInPrepareHeader(IntPtr hWaveIn, ref WaveHdr lpWaveInHdr, UInt32 uSize);
[DllImport("winmm.dll")]
//标记为不可用的缓冲区
public static extern int waveInUnprepareHeader(IntPtr hWaveIn, ref WaveHdr lpWaveInHdr, UInt32 uSize);
[DllImport("winmm.dll")]
//把缓冲区内容重置
public static extern int waveInReset(IntPtr hwi);
[DllImport("winmm.dll")]
//开始录制
public static extern int waveInStart(IntPtr hwi);
[DllImport("winmm.dll")]
//停止录制
public static extern int waveInStop(IntPtr hwi);

然后 我们要把接收到的波形数据放入到一个缓冲区里面

[StructLayout(LayoutKind.Sequential)]
//接受的波形数据放入的缓冲区
public struct WaveHdr
{

public IntPtr lpData;//缓冲区
public UInt32 dwBufferLength;//缓冲区长度
public UInt32 dwBytesRecorded;//某一刻读取到了多少字节的数据
public UInt32 dwUser;//自定义数据
public UInt32 dwFlags;
public UInt32 dwLoops;//是否循环
public IntPtr lpNext;//链表的下一缓冲区
public UInt32 reserved;//没实际意义
}

[StructLayout(LayoutKind.Sequential)]
//波形格式
public struct WaveFormatEx
{
public UInt16 wFormatTag;//波形的类型
public UInt16 nChannels;//通道数（1，单声道   2，立体音）
public UInt32 nSamplesPerSec;//采样率
public UInt32 nAvgBytesPerSec;//字节率
public UInt16 nBlockAlign;
public UInt16 wBitsPerSample;//每个样多少位
public UInt16 cbSize;//长度
}

但是在这里我们需要一个delegate的委托事件，其作用是在缓冲区满了或者waveinopen和waveinclose的时候被调用。

public delegate void WaveDelegate(IntPtr hwi, UInt32 uMsg, UInt32 dwInstance, UInt32 dwParam1, UInt32 dwParam2);

上传到百度识别的接口

在全部获取到麦克风语音接收的信息之后，我们需要把识别的波形上传到百度识别的接口上，在这里我们就用HTTP协议来将我们获得的东西上传上去

/// <summary>
/// 通过HTTP协议去上传base64数据
/// </summary>
/// <param name="URL">服务器的url</param>
/// <param name="strPostdata">上传的东西</param>
/// <param name="strEncoding">采用的编码格式</param>
/// <returns></returns>
public static string OpenReadWithHttps(string URL, string strPostdata, string strEncoding)
{
Encoding encoding = Encoding.Default;
//默认的编码格式为default（GB2312）
HttpWebRequest request = (HttpWebRequest)WebRequest.Create(URL);
//向自定义的URL链接发送请求 request
request.Method = "post";
//请求的方式为post
request.Accept = "*/*";
//告诉服务器能接受*/*（任意）的参数类型
request.ContentType = "application/x-www-form-urlencoded";
//最常见的post提交数据的方式
byte[] buffer = encoding.GetBytes(strPostdata);
//用一个byte数组接收发送的数据字节
request.ContentLength = buffer.Length;
//告诉服务器自己上传的数组长度
request.GetRequestStream().Write(buffer, 0, buffer.Length);
//写入请求流从第一位开始写入buffer数组，写入长度为buffer.Length的数据流
HttpWebResponse response = (HttpWebResponse)request.GetResponse();
//从服务器得到的数据为请求获得的数据
using (StreamReader reader = new StreamReader(response.GetResponseStream(), Encoding.GetEncoding(strEncoding)))
{
//返回从URL获得的内容信息
return reader.ReadToEnd();
}
}

判断是否在录入音频

每个语音都是一个缓冲区，等缓冲区满了，要提供新缓冲区，等缓冲区满了，要提供新缓冲区

static void waveInHandler(IntPtr hwi, UInt32 uMsg, UInt32 dwInstance, UInt32 dwParam1, UInt32 dwParam2)
{
switch (uMsg)
{
case 0x3BE: break;
case 0x3C0:
unsafe
{
var waveHdr = (WaveHdr*)dwParam1;
}
break;
case 0x3BF: break;
}
}

Main函数总结

小的模块说的差不多了，下面就从Main函数说起，中间还会穿插一些小的模块

首先我们设置波纹的格式

static void Main(string[] args)
{
try
{
var inputFormat = new WaveFormatEx();//波形格式
inputFormat.wFormatTag = 1;//波形类型
inputFormat.nChannels = 1;
inputFormat.nSamplesPerSec = 8000;
inputFormat.nAvgBytesPerSec = 16000;
inputFormat.nBlockAlign = 2;
inputFormat.wBitsPerSample = 16;
inputFormat.cbSize = 0;

由于我们是语音识别不是就识别一次，所以我们下面要进入一个死循环

这里我们在等候语音的输入

for (;;)
{

waveInOpen(out inputDevice, UInt32.MaxValue, ref inputFormat, new WaveDelegate(waveInHandler), 0, 0x00030000);

int bufferSize = 960000;
var buffer1 = new WaveHdr();
buffer1.lpData = Marshal.AllocHGlobal(bufferSize);
buffer1.dwBufferLength = (UInt32)bufferSize;
buffer1.dwLoops = 1;
waveInPrepareHeader(inputDevice, ref buffer1, (UInt32)Marshal.SizeOf(typeof(WaveHdr)));
waveInAddBuffer(inputDevice, ref buffer1, (UInt32)Marshal.SizeOf(typeof(WaveHdr)));

SpeechRecognitionEngine recognizer = null;
foreach (var installed in SpeechRecognitionEngine.InstalledRecognizers())
{
if (installed.Culture.Name.Equals("zh-CN", StringComparison.CurrentCultureIgnoreCase) && installed.Id.Equals("MS-2052-80-DESK"))
{
recognizer = new SpeechRecognitionEngine(installed);
break;
}
}
var grammars = new GrammarBuilder();
grammars.AppendDictation();
recognizer.LoadGrammar(new Grammar(grammars));
recognizer.SetInputToDefaultAudioDevice();

bool recognizeStarted = false;
int speechCount = 0;
int silenceCount = 0;

Console.WriteLine("正在等候语音输入...");
recognizer.RecognizeAsync(RecognizeMode.Multiple);
waveInStart(inputDevice);

当说话的时候开始分析语音

for (;;)
{
if (!recognizeStarted)
{
if (recognizer.AudioState == AudioState.Speech)
speechCount++;
else speechCount = 0;
}
if (!recognizeStarted && speechCount >= 2)
{
recognizeStarted = true;
speechCount = 0;
Console.WriteLine("检测到语音输入，正在录制...");
}
if (recognizeStarted)
{
if (recognizer.AudioState == AudioState.Silence)
silenceCount++;
else silenceCount = 0;
}
//checkingMutex.Set();
if (recognizeStarted && silenceCount >= 220)
{
//checkingMutex.Reset();
silenceCount = 0;
unsafe
{
Console.WriteLine("正在分析语音数据...");

waveInReset(inputDevice);
waveInStop(inputDevice);
recognizer.RecognizeAsyncStop();

在上面的代码中有判断环境噪音的代码

if (recognizer.AudioState == AudioState.Silence)
silenceCount++;
else silenceCount = 0;

silenceCount 就是统计静音状态持续了多久，到了一定值，就可以发送语音到识别平台了

然后我们就要用到百度给予的接口和key了

var apiKey = "百key";
var secretKey = "百度给的密码key";
var token = OpenReadWithHttps("百度给你提供的API接口地址http" + $"?grant_type={ "client_credentials" }&client_id={ apiKey }&client_secret={ secretKey }", String.Empty, "utf-8");
var tokenPrefix = "\"access_token\":[\"";
int i;
token = token.Substring(i = token.IndexOf(tokenPrefix) + tokenPrefix.Length + 1, token.IndexOf("\"", i + tokenPrefix.Length) - i);

var postData = new StringBuilder();
postData.Append("{").Append($"\"format\":\"pcm\",\"rate\":8000,\"channel\":1,\"token\":\"{ token }\",\"cuid\":\"F96625D0-0FBC-491C-B617-9EC0B3A0D5A6\",\"lan\":\"en\",");
var base64Data = new byte[buffer1.dwBytesRecorded];
Marshal.Copy(buffer1.lpData, base64Data, 0, (int)buffer1.dwBytesRecorded);
var base64 = Convert.ToBase64String(base64Data);
postData.Append("\"speech\":\"").Append(base64).Append("\",").Append($"\"len\":{ buffer1.dwBytesRecorded }").Append("}");
try
{
Console.Write("\n识别结果: ");
Marshal.FreeHGlobal(buffer1.lpData);
var result = OpenReadWithHttps("http://vop.baidu.com/server_api", postData.ToString(), "utf-8");
var prefix = "\"result\":[\"";
result = result.Substring(i = result.IndexOf(prefix) + prefix.Length, result.LastIndexOf("\"]") - i + 1);
string[] restt = result.Split('\"');
var restlt = restt[0];
Console.WriteLine(restlt);
//string resultfinally = Recognize(restlt);
try
{
string resultfinally = Recognize(restlt);
loading(resultfinally, "word.txt");
}
catch (Exception ex)
{
Console.WriteLine(ex.Message);
//(new SpVoiceClass()).Speak("你说的有些不标准，请重新说");
}
Console.WriteLine();
}
catch (Exception ex)
{
Console.WriteLine("无法识别所说的话语。\n");

}
//checkingMutex.Set();
}
//checking.Dispose(checkingFinished);
break;
}
Thread.Sleep(1);
}

//checkingFinished.WaitOne();
}
}
catch (Exception exception)
{
Console.WriteLine(exception);
}

}
}
}

在这里你会发现，我做了一个语音识别和回复，识别目录下的文档里的内容，然后对比，对比到以后将下一句转换为语音。需要用到两个自定义的函数

第一个是判断你说的话是否是在给定的文本里面

public static void loading(string listen, string url)
{
var file = File.OpenRead(url);
var sr = new StreamReader(file);
List<string> include = new List<string>();
while (!sr.EndOfStream)
{
var str = sr.ReadLine();
foreach (var chara in str)
if (!char.IsLetter(chara))
str = str.Replace(chara, ' ');
str = str.Trim();
include.Add(str);
}
for (int i = 0; i < include.Count; i++)
{
if (String.Compare(listen.Trim(), include[i].Trim(), StringComparison.CurrentCultureIgnoreCase) == 0)
{
SpeechSynthesizer speaker = new SpeechSynthesizer();
speaker.SetOutputToDefaultAudioDevice();
speaker.Speak(include[i + 1]);
return;
}
}
SpeechSynthesizer speak = new SpeechSynthesizer();
speak.SetOutputToDefaultAudioDevice();
speak.Speak("口音有问题，请重说。");
throw new Exception("口音有问题，请重说。");
}

第二个是判断是否跟自定义的语句匹配并说出下一句

public static string Recognize(string getin)
{
var responses = new string[]
{
"楼主帅吗",
"当然了",
"聪明吗",
"必须的",
//你想写和你想输出的语句
};
getin = getin.ToLower();
foreach (var chara in getin)
if (!char.IsLetter(chara))
getin = getin.Replace(chara, ' ');
getin = getin.Trim();
int matches;
var k = getin.Split();
for(var i = 0; i < responses.Length; i++)
{
responses[i] = responses[i].ToLower();
foreach (var chara in responses[i])
if (!char.IsLetter(chara))
responses[i] = responses[i].Replace(chara, ' ');
responses[i] = responses[i].Trim();
}
foreach (var repWord in responses)
{
matches = 0;
var j = repWord.Split();
foreach (var myword in k)
{
if (j.Contains(myword))
{
matches++;
if (((float)matches / j.Length) >= 0.5F)
return repWord;
}
}
}
return "你说错了，请重说";
}

这里还是有一个小问题，就是你说的语句返识别返还回来会有标点符号，这里我们就把符号全部给抛弃了

我这边做的是英语的语音识别，在发送的json串的时候最后的len用的是en,在语种选择的时候是不区分大小写的，但是好像只支持三种默认中文（zh）。 中文=zh、粤语=ct、英文=en。

总的来说就这些东西，如果有什么疑问和建议或者纠正，可以直接告诉我，期待大神们的指点。

此外。在此特别感谢给我这个程序最大的技术支持的人。我们群里的大佬RURI（也叫Azure）。