基于Accord.Audio和百度语言识别

目标需求

使用录音形式，模拟微信语音聊天。按住录音，松开发送语音，并完成语音识别。

ps：百度的语言识别有60秒长度限制，需要自己做好控制。

实现方案

采用C# winform 程序实现桌面版，采用Accord 实现语音录制停止等基础语音操作，操作停止按钮，

自动调用百度语言识别接口将识别内容显示在文本框中。

备注，语音识别需要配套阵列麦克风，（请先注册百度开发者）百度语音识别接口请参考：http://ai.baidu.com/docs#/ASR-Online-Csharp-SDK/top

实现效果展示

实现过程

1、下载Accord 完成语音操作引用

accord 官方地址：http://accord-framework.net/intro.html

官网中有示例demo，笔者的就是在示例demo上做改造的。

建立自己的项目，引用包中的dll

界面代码：

using System;

using System.Drawing;

using System.IO;

using System.Windows.Forms;

using Accord.Audio;

using Accord.Audio.Formats;

using Accord.DirectSound;

using Accord.Audio.Filters;

using Baidu.Aip.API;

namespace SampleApp

{

public partial class MainForm : Form

{

private MemoryStream stream;

private IAudioSource source;

private IAudioOutput output;

private WaveEncoder encoder;

private WaveDecoder decoder;

private float[] current;

private int frames;

private int samples;

private TimeSpan duration;

/// <summary>

/// 备注，语音识别需要配套阵列麦克风

/// </summary>

public MainForm()

{

InitializeComponent();

// Configure the wavechart

chart.SimpleMode = true;

chart.AddWaveform("wave", Color.Green, 1, false);

updateButtons();

// Application.Idle += ProcessFrame;

}

void ProcessFrame(object sender, EventArgs e) {

}

/// <summary>

/// 从声卡开始录制音频

/// </summary>

///

private void btnRecord_Click(object sender, EventArgs e)

{

// Create capture device

source = new AudioCaptureDevice()//这里是核心

{

// Listen on 22050 Hz

DesiredFrameSize = 4096,

SampleRate = 16000,//采样率

//SampleRate = 22050,//采样率

Channels=1,

// We will be reading 16-bit PCM

Format = SampleFormat.Format16Bit

};

// Wire up some events

source.NewFrame += source_NewFrame;

source.AudioSourceError += source_AudioSourceError;

// Create buffer for wavechart control

current = new float[source.DesiredFrameSize];

// Create stream to store file

stream = new MemoryStream();

encoder = new WaveEncoder(stream);

// Start

source.Start();

updateButtons();

}

/// <summary>

/// 播放录制的音频流。

/// </summary>

///

private void btnPlay_Click(object sender, EventArgs e)

{

// First, we rewind the stream

stream.Seek(0, SeekOrigin.Begin);

// Then we create a decoder for it

decoder = new WaveDecoder(stream);

// Configure the track bar so the cursor

// can show the proper current position

if (trackBar1.Value < decoder.Frames)

decoder.Seek(trackBar1.Value);

trackBar1.Maximum = decoder.Samples;

// Here we can create the output audio device that will be playing the recording

output = new AudioOutputDevice(this.Handle, decoder.SampleRate, decoder.Channels);

// Wire up some events

output.FramePlayingStarted += output_FramePlayingStarted;

output.NewFrameRequested += output_NewFrameRequested;

output.Stopped += output_PlayingFinished;

// Start playing!

output.Play();

updateButtons();

}

/// <summary>

/// 停止录制或播放流。

/// </summary>

///

private void btnStop_Click(object sender, EventArgs e)

{

// Stops both cases

if (source != null)

{

// If we were recording

source.SignalToStop();

source.WaitForStop();

}

if (output != null)

{

// If we were playing

output.SignalToStop();

output.WaitForStop();

}

updateButtons();

// Also zero out the buffers and screen

Array.Clear(current, 0, current.Length);

updateWaveform(current, current.Length);

SpeechAPI speechApi = new SpeechAPI();

string result = speechApi.AsrData(stream,"wav");

tb_result.Text = "语音识别结果："+result;

}

/// <summary>

/// 当音频有错误时，将调用这个回调函数。

///

/// </summary>

///

private void source_AudioSourceError(object sender, AudioSourceErrorEventArgs e)

{

throw new Exception(e.Description);

}

/// <summary>

///

/// 每当有新的输入音频帧时，该方法将被调用。

///

/// </summary>

///

private void source_NewFrame(object sender, NewFrameEventArgs eventArgs)

{

eventArgs.Signal.CopyTo(current);

updateWaveform(current, eventArgs.Signal.Length);

encoder.Encode(eventArgs.Signal);

duration += eventArgs.Signal.Duration;

samples += eventArgs.Signal.Samples;

frames += eventArgs.Signal.Length;

}

private void output_FramePlayingStarted(object sender, PlayFrameEventArgs e)

{

updateTrackbar(e.FrameIndex);

if (e.FrameIndex + e.Count < decoder.Frames)

{

int previous = decoder.Position;

decoder.Seek(e.FrameIndex);

Signal s = decoder.Decode(e.Count);

decoder.Seek(previous);

updateWaveform(s.ToFloat(), s.Length);

}

private void output_PlayingFinished(object sender, EventArgs e)

{

updateButtons();

Array.Clear(current, 0, current.Length);

updateWaveform(current, current.Length);

}

///

private void output_NewFrameRequested(object sender, NewFrameRequestedEventArgs e)

{

e.FrameIndex = decoder.Position;

Signal signal = decoder.Decode(e.Frames);

if (signal == null)

{

e.Stop = true;

return;

}

e.Frames = signal.Length;

signal.CopyTo(e.Buffer);

}

private void updateWaveform(float[] samples, int length)

{

if (InvokeRequired)

{

BeginInvoke(new Action(() =>

{

chart.UpdateWaveform("wave", samples, length);

}));

}

else

{

chart.UpdateWaveform("wave", current, length);

}

///

private void updateTrackbar(int value)

{

if (InvokeRequired)

{

BeginInvoke(new Action(() =>

{

trackBar1.Value = Math.Max(trackBar1.Minimum, Math.Min(trackBar1.Maximum, value));

}));

}

else

{

trackBar1.Value = Math.Max(trackBar1.Minimum, Math.Min(trackBar1.Maximum, value));

}

private void updateButtons()

{

if (InvokeRequired)

{

BeginInvoke(new Action(updateButtons));

return;

}

if (source != null && source.IsRunning)

{

btnBwd.Enabled = false;

btnFwd.Enabled = false;

btnPlay.Enabled = false;

btnStop.Enabled = true;

btnRecord.Enabled = false;

trackBar1.Enabled = false;

}

else if (output != null && output.IsRunning)

{

btnBwd.Enabled = false;

btnFwd.Enabled = false;

btnPlay.Enabled = false;

btnStop.Enabled = true;

btnRecord.Enabled = false;

trackBar1.Enabled = true;

}

else

{

btnBwd.Enabled = false;

btnFwd.Enabled = false;

btnPlay.Enabled = stream != null;

btnStop.Enabled = false;

btnRecord.Enabled = true;

trackBar1.Enabled = decoder != null;

trackBar1.Value = 0;

}

private void MainFormFormClosed(object sender, FormClosedEventArgs e)

{

if (source != null) source.SignalToStop();

if (output != null) output.SignalToStop();

}

private void saveFileDialog1_FileOk(object sender, System.ComponentModel.CancelEventArgs e)

{

Stream fileStream = saveFileDialog1.OpenFile();

stream.WriteTo(fileStream);

fileStream.Close();

}

private void saveToolStripMenuItem_Click(object sender, EventArgs e)

{

saveFileDialog1.ShowDialog(this);

}

private void updateTimer_Tick(object sender, EventArgs e)

{

lbLength.Text = String.Format("Length: {0:00.00} sec.", duration.Seconds);

}

private void aboutToolStripMenuItem_Click(object sender, EventArgs e)

{

new AboutBox().ShowDialog(this);

}

private void closeToolStripMenuItem_Click(object sender, EventArgs e)

{

Close();

}

private void btnIncreaseVolume_Click(object sender, EventArgs e)

{

adjustVolume(1.25f);

}

private void btnDecreaseVolume_Click(object sender, EventArgs e)

{

adjustVolume(0.75f);

}

private void adjustVolume(float value)

{

stream.Seek(0, SeekOrigin.Begin);

decoder = new WaveDecoder(stream);

var signal = decoder.Decode();

var volume = new VolumeFilter(value);

volume.ApplyInPlace(signal);

stream.Seek(0, SeekOrigin.Begin);

encoder = new WaveEncoder(stream);

encoder.Encode(signal);

}

百度语音识别接口

百度已经提供sdk，对于支持语音格式如下。

支持的语音格式

原始 PCM 的录音参数必须符合 8k/16k 采样率、16bit 位深、单声道，支持的格式有：pcm（不压缩）、wav（不压缩，pcm编码）、amr（压缩格式）。

        public string AsrData(string filePath, string format = "pcm", int rate = 16000){         
                var data =File.ReadAllBytes(filePath);         
                var result = _asrClient.Recognize(data, format, 16000);      
                return result.ToString();}