Performance and Testing¶
Why is it difficult to recognize when speaking faster?¶
When speaking faster, people’s pronunciation often changes. If the speaking speed is too fast, the pronunciation may and is easy to be distorted, leading to changes in voice characteristics. Like people talking, if the speaker speaks very fast, it may be difficult for the listener to understand what the speaker is saying. Therefore, the recognition rate will be reduced, especially in noisy conditions.
At present, our company has corresponding technical solutions, which can maintain a high recognition rate under normal fast speech speed. For more information, please consult our technical support.
What aspects should be paid attention to during product testing?¶
During the product test, please pay attention to avoid people talking around during the test. At the same time, it should be noted that there should be no noise in the test environment except for the noise required by the test. Large echo shall be avoided in the test room to reduce reverberation. During the whole machine test, attention shall be paid to avoid the noise of the machine itself interfering with the test.
At the same time, it is better to use our test standards for the test. Pay attention to the voice speed of the tester and whether the microphone adopts the specifications recommended by our company to achieve good test results.
After the whole machine test, it is found that there is a problem with the recognition rate. What can be checked?¶
In case of this problem, please first check the overall structure, especially whether the microphone is opened and installed according to our suggestions, whether the back of the microphone is sealed and fixed, and whether it is far away from the noise source. If there is no problem, you can try to take out the voice module and microphone and test them in the same environment to see if the recognition rate is improved? If the bare board test is OK, please check the structure again. If the bare board test effect is not good, please analyze the bare board to improve the identification rate.
After the voice module test, there is a problem with the recognition rate. What can I check?¶
If the module provided by our company is used, please check whether the command word optimization is not good, and then modify the software. For circuit boards designed by developers themselves, we can burn our standard modules to the same firmware, and select the same microphone for comparative testing in the same environment. If the identification of our standard modules is normal, it can be considered that it is the problem of the circuit board designed by ourselves. Please check according to the relevant requirements of our hardware design. If you cannot locate the problem, you can contact our technical support to assist in the analysis.
In addition, during the test, it is recommended to take videos on the test site and use your mobile phone to record synchronously at the microphone position of the module. Our technical support is requested to send the videos and recordings to us synchronously during the analysis, which can speed up the analysis and processing.
Is there a convenient recording method for collecting test audio?¶
You can use a mobile phone to record. When recording, the microphone of the mobile phone and the microphone of the voice module should be in the same direction, and the mobile phone should be placed next to the voice module (the mobile phone should avoid the impact of other equipment vibration or air outlet). Then the mobile phone can start recording, and record while testing according to the actual test method of the product.
What are the requirements for the environment when testing?¶
In order to achieve better test results, it is recommended to build test equipment and use the product equipment models recommended by our company, such as hi fi audio. The microphone used by the module is - 32 ± 3dB. Test room size: a room not less than 4m * 4m and not more than 6m * 6m, used to simulate the home environment.
The test room shall be soundproofed or kept relatively independent, and the external sound and the test environment shall not affect each other. There is no obvious noise outside the test room (such as car and vegetable market), and the reverberation value range is 0.3-0.6. Before the test, it is better to confirm the background noise. It is recommended that the quiet environment be kept at 35-45dB, and the news noise range be 58-60dB. In a quiet environment, the test voice or broadcast voice shall be kept at about 60dB, and in a noise environment, the test voice or broadcast voice shall be kept at 70-75dB, and the signal-to-noise ratio shall be>15dB.
In order to ensure the accuracy of the test results, it is recommended to use more than or equal to 2 test modules in each group, and average the test results.
What are the precautions when using your company’s automated test?¶
Our company provides automated testing tools, and the relevant usage methods can be found in this document center. This tool only counts the recognition rate/error recognition rate of command terms. In order to ensure the preparation of recognition, the automatic test firmware must turn off the printing of other debugging serial ports except the recognition printing. If the firmware is made with broadcast function, the audio broadcast time cannot exceed 2s, otherwise the overall identification result will be affected. If the produced firmware can only be used for word recognition after wake-up, it is not suitable to use automated testing tools, and manual testing is recommended.
The automatic test firmware must use a single network, and the audio of the automatic test must use the standardized audio file. In order to improve the test efficiency, it is recommended to remove the mute before and after the automatic test audio file. During the automatic test, HUB must be powered by a separate power supply, and an artificial mouth or a high fidelity speaker must be used to play audio files. The signal-to-noise ratio in quiet and noisy environments is>15dB.
What are the precautions when using manual testing?¶
If manual testing is adopted, it is recommended that the tester should be aged between 18 and 60 (except for children’s products), and use standard Mandarin to read command words. The tester should not read the command words too fast, and the speed should be controlled at 150-180 words/minute in Chinese Putonghua. In a quiet environment, the voice or broadcast voice shall be kept at about 60dB; in a noisy environment, the voice or broadcast voice shall be kept at 70-75dB, and the signal-to-noise ratio shall be>15dB. The test distance is 3m-5m. During the test, the loudspeakers for playing noise should not be placed directly against the microphone, but in the same direction or back, to avoid the noise waves directly interfering with the waveform from the human voice to the microphone.