c# - 使用Microsoft SpeechRecognitionEngine时如何提高结果的准确性？

Question

我目前正在开发一个 DialerService 项目。其中一项功能是将录制的 .wav 媒体文件解释为纯文本。我使用 SpeechRecognitionEngine 尝试解释内容，但得到了一些不准确的结果，有时甚至是毫无意义的断句。

.wav 文件是两个或多个客户之间的电话交谈的录音文件，我测试的文件是我与同事进行的非常简单和简短的对话。

所以我的问题是我怎样才能提高解释的准确性以及为此目的如何改进我的代码？我知道添加语法将有助于识别一些关键字，但我需要的是对我从用户那里记录的内容进行大体解释。

下面是我的工作代码:

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.IO;
using System.Threading.Tasks;
using System.Speech.Recognition;
using System.Speech.AudioFormat;
using System.Web;

namespace VoiceRecognition
{
    class Program
    {

        static bool completed;

        static void Main(string[] args)
        {
            using (
             SpeechRecognitionEngine recognizer =
                    new SpeechRecognitionEngine(
                        new System.Globalization.CultureInfo("en-US")))
            {

                // Create and load a grammar.
                Grammar dictation = new DictationGrammar();
                dictation.Name = "Dictation Grammar";

                recognizer.LoadGrammar(new DictationGrammar());

                recognizer.SetInputToWaveFile(@"C:\Projects2\VoiceRecognition2\conf_with_vincent_1.wav");
                // Attach event handlers for the results of recognition.
                //recognizer.AudioLevelUpdated += new EventHandler<AudioLevelUpdatedEventArgs>(recognizer_AudioLevelUpdated);
                //recognizer.AudioStateChanged += new EventHandler<AudioStateChangedEventArgs>(recognizer_AudioStateChanged);

                recognizer.SpeechRecognized  +=  new EventHandler<SpeechRecognizedEventArgs>(recognizer_SpeechRecognized);
                recognizer.RecognizeCompleted += new EventHandler<RecognizeCompletedEventArgs>(recognizer_RecognizeCompleted);

                // Perform recognition on the entire file.
                Console.WriteLine("Starting asynchronous recognition...");
                completed = false;
                //recognizer.RecognizeAsync();
                recognizer.RecognizeAsync(RecognizeMode.Multiple);

                // Keep the console window open.
                while (!completed)
                {
                    Console.ReadLine();
                }
                Console.WriteLine("Done.");
            }

            Console.WriteLine();
            Console.WriteLine("Press any key to exit...");
            Console.ReadKey();


        }

        // Handle the Audio state event.
        static void recognizer_AudioStateChanged(object sender, AudioStateChangedEventArgs e)
        {
            Console.WriteLine("The new audio state is: " + e.AudioState);
        }

        static void recognizer_AudioLevelUpdated(object sender, AudioLevelUpdatedEventArgs e)
        {
            Console.WriteLine("The audio level is now: {0}.", e.AudioLevel);
        }


        // Handle the SpeechRecognized event.
        static void recognizer_SpeechRecognized(object sender, SpeechRecognizedEventArgs e)
        {
            if (e.Result != null && e.Result.Text != null)
            {
                Console.WriteLine("  Recognized text =  {0}", e.Result.Text);
            }
            else
            {
                Console.WriteLine("  Recognized text not available.");
            }
        }

        // Handle the RecognizeCompleted event.
        static void recognizer_RecognizeCompleted(object sender, RecognizeCompletedEventArgs e)
        {
            if (e.Error != null)
            {
                Console.WriteLine("  Error encountered, {0}: {1}",
                e.Error.GetType().Name, e.Error.Message);
            }
            if (e.Cancelled)
            {
                Console.WriteLine("  Operation cancelled.");
            }
            if (e.InputStreamEnded)
            {
                Console.WriteLine("  End of stream encountered.");
            }
            completed = true;
        }



    }
}

另一个类是:

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Speech.Recognition;

public class SpeechReconizer
{

    SpeechRecognitionEngine _speechRecognitionEngine;
    public SpeechReconitionResult ReadResult { get; set; }

    public SpeechReconizer()
    {
        Grammar dictation = new DictationGrammar();
        dictation.Name = "Dictation Grammar";



        _speechRecognitionEngine = new SpeechRecognitionEngine();
        _speechRecognitionEngine.SetInputToDefaultAudioDevice();
        _speechRecognitionEngine.LoadGrammar(dictation);
        _speechRecognitionEngine.InitialSilenceTimeout = TimeSpan.FromSeconds(3);
        _speechRecognitionEngine.BabbleTimeout = TimeSpan.FromSeconds(2);
        _speechRecognitionEngine.EndSilenceTimeout = TimeSpan.FromSeconds(1);
        _speechRecognitionEngine.EndSilenceTimeoutAmbiguous = TimeSpan.FromSeconds(1.5);
        _speechRecognitionEngine.RecognizeAsync(RecognizeMode.Multiple);
        _speechRecognitionEngine.SpeechRecognized += RecognizerSpeechRecognized;
        _speechRecognitionEngine.RecognizeCompleted += RecognizerRecognizeCompleted;
    }



    public SpeechReconitionResult ReadSpeech(string sourceAudio)
    {
        ReadResult = new SpeechReconitionResult();

        _speechRecognitionEngine.SetInputToWaveFile(sourceAudio);


        _speechRecognitionEngine.Recognize();
        return ReadResult;

    }

    private void RecognizerSpeechRecognized(object sender, SpeechRecognizedEventArgs e)
    {
        if (e.Result != null && e.Result.Text != null)
        {
            ReadResult.Success = true;
            ReadResult.Text = e.Result.Text;
        }
        else
        {
            ReadResult.Text = "Recognized text not available.";
        }
    }

    private void RecognizerRecognizeCompleted(object sender, RecognizeCompletedEventArgs e)
    {
        if (e.Error != null)
        {
            ReadResult.Success = false;
            ReadResult.ErrorMessage = string.Format("{0}: {1}",
                          e.Error.GetType().Name, e.Error.Message);
        }
        if (e.Cancelled)
        {
            ReadResult.Success = false;
            ReadResult.ErrorMessage = "Operation cancelled.";
        }
    }

}
public class SpeechReconitionResult
{
    public string Text { get; set; }
    public bool Success { get; set; }
    public string ErrorMessage { get; set; }
    public bool Complete { get; set; }
}

测试结果是(在控制台):

Starting asynchronous recognition...
  Recognized text = Helence and the globe or east
  Recognized text = alarmed
  Recognized text = and client thanks
  Recognized text = what aren't going to do and that they
  Recognized text = aren't goint to rule
  Recognized text = working to dear E
  Recognized text = N
  Recognized text = at dinner
  Recognized text = and
  Recognized text = that going there
  Recognized text = and you have a 98 no problem bars
  End of stream encountered.

实际内容是什么:- 你好文森特。- 你好，鲍里斯。-你好吗？-我没事。-你今天打算做什么？-我要看电视，吃晚饭然后回家。-感谢您有一个愉快的一天。-没问题。

Answer 1

System.Speech.Recognition 支持默认的 Windows 语音识别。它是为单个用户设计的，可以由用户通过windows语音识别训练进行训练。

您可能需要的是专为低质量音频设计的 Microsoft.Speech.Recognition 库。它的工作方式几乎相同，但是，它不是为听写而设计的。它更多地用于检测来自电话质量音频的命令。如果您想试一试，我发现的最新版本在这里:http://www.microsoft.com/en-us/download/details.aspx?id=27226