Shogo Ando

Hello everyone.

This is a series of articles introducing tips for improving the accuracy of speech recognition. In this fourth article, we will look at noise, a formidable enemy of speech recognition. Needless to say, the accuracy of speech recognition decreases in noisy environments. However, in actual situations where speech recognition is used, there are many situations where ambient noise is unavoidable.

In this article, we will take a closer look at how noise affects speech recognition and introduce specific techniques to improve the accuracy of speech recognition even in noisy environments.

The effect of noise on speech recognition

Now, I think many people are wondering, "How much noise is acceptable?" It's true that humans can often hear the other person's voice even if the surroundings are somewhat noisy. Similarly, voice recognition can often recognize voices without any problems if the noise level is mild. However, when you actually measure the recognition rate, you will find that the rate drops significantly as the noise level increases, and even small noise that is not very noticeable can affect the recognition rate.

For this reason, it is important to minimize noise as much as possible in order to maximize the accuracy of speech recognition.

The waveforms below represent, from left to right, audio with "little noise," "a lot of noise," and "extremely noisy."

• "Low noise" (left)
  This is an ideal waveform. The audio waveform is divided into parts where it fluctuates (speech) and parts where it does not fluctuate at all (silence). There is almost no noise, and this results in the highest accuracy of voice recognition.
• "There is a lot of noise" (center)
  This shows a situation where there is a little noise. The audio waveform fluctuates even in places where there should be no sound, and you can see that noise is mixed in. With this level of noise, the accuracy of voice recognition will visibly decrease.
• "There's a lot of noise" (right)
  The entire waveform fluctuates greatly, making it difficult to visually determine which parts are being spoken. This type of speech significantly reduces recognition accuracy, and appropriate countermeasures are required.

The answer to the question, "How much does noise affect the recognition rate?" varies depending on the situation, so it is difficult to give a general answer. However, if there is noise on the level shown in the center waveform above, you should be aware of the potential for a decrease in recognition accuracy.

What to do in a noisy environment?

Now, let's explain what measures you should take when there is a lot of noise. There are three main measures you can take.

• Turn down the microphone volume, place your mouth close to the microphone, and speak loudly
• Use a directional microphone
• Use noise reduction(It can be effective, but you need to be careful.)

The most effective measure is the first one mentioned: "Turn down the microphone volume, put your mouth close to the microphone, and speak loudly." This is like "speaking loudly close to the other person's ear" when talking to someone. In noisy environments, this is the most effective way to improve the accuracy of voice recognition.

Using the waveforms above, the waveform on the left contains a lot of noise. By placing your mouth close to the microphone and speaking loudly, you can make your speech louder relative to the noise, as shown in the waveform in the middle. However, this alone increases the likelihood that the sound will be too loud and cause distortion. Therefore, by turning down the microphone volume, you can make the volume appropriate, as shown in the waveform on the right. This increases the likelihood that the accuracy of speech recognition will improve.

Furthermore, Use a directional microphoneDirectional microphones also help to effectively pick up speech in a noisy environment. Directional microphones focus on picking up sound from a specific direction, reducing background noise and making speech sound louder.

In addition, Use noise reduction (noise cancellation) functionsThis is also an effective measure. However, it is important to note that some noise reduction methods may change the characteristics of sounds recorded with a normal microphone, which may result in a decrease in the accuracy of speech recognition. When using noise reduction methods, please be sure to consider these risks and use them appropriately.

Other noise countermeasures

In addition to this, you can gradually improve the accuracy of voice recognition by considering the following measures. There are few measures that have an immediate effect, but by keeping these points in mind, you will be able to minimize the impact of noise.

• Stay as far away from the noise source as possible
• Avoid speaking when there is noise(Example: When a train passes on the station platform)
• Speak in a place where noise is less likely to reach you, such as in the shade

Summary

This time, we introduced the effect that noise has on the accuracy of speech recognition and techniques to suppress noise as much as possible. If you are having trouble using a speech recognition system due to too much noise, we recommend trying the "Turn down the microphone volume, place your mouth close to the microphone, and speak loudlyPlease try the following method.

This series, which has introduced tips for improving the accuracy of voice recognition, will conclude with the next installment.Use in an appropriate communication environment"We will explain about this.

The accuracy of speech recognition (speech recognition rate) can be improved by the way you speak. What are the tips for how to speak?

How to Choose the Right Microphone - Speech recognition rate can change depending on the microphone!

Improve your speech recognition accuracy! How to properly use and set up your microphone and apps

Is the Internet to blame for voice recognition not working? How to create a communication environment that allows voice recognition to work smoothly