An automated speech engine effect testing system and method thereof

Abstract

The invention discloses an automatic voice engine effect testing system and method. The automatic voice engine effect testing system comprises a corpus playing device, a corpus playing distance adjusting device, a noise playing device, a noise playing adjusting device and an angle adjusting device. The distance switching between the pickup front end and the corpus playing device is realized, the switching of simulated near-field and far-field scenes is completed, the accurate positioning is realized by using the contact switch, the consistency of each scene is improved, different noise anglescan be tested in each scene, the acquired data is more complete, and the efficiency is higher. The upper computer is used for controlling; the audio is uploaded to the cloud end in stages, the audio collected by the front end is divided into clips according to scenes, data is sent to the cloud end to be processed every time one scene is completed, local space is released, recording and processingare achieved at the same time, invalid data caused by errors can be effectively prevented, quasi-real-time result output is achieved, and the whole test period is greatly shortened.

Description

technical field [0001] The invention relates to the field of speech recognition, in particular to an automatic speech engine effect testing system and a method thereof. Background technique [0002] At present, the testing of the speech engine is mainly manual testing and semi-automated testing; [0003] Manual testing is generally to experience the effect of the engine artificially, which is convenient and fast, and the intuitive feeling is close to the user experience, but the data of the test sample is very small, the consistency is difficult to control, and the test results are not very convincing; [0004] If the recording method is used to switch the scene for multiple tests, the sample data can be improved, but due to multiple manual interventions, it is easy to cause multiple subtle errors, and seriously damage the continuity of the test and lengthen the test cycle; [0005] In the test of using the server to read the audio running algorithm, the above-mentioned pro...

AI Technical Summary

Problems solved by technology

[0003] Manual testing is generally to experience the effect of the engine artificially, which is convenient and fast, and the intuitive feeling is close to the user experience, but the data of the test sample is very small, the consistency is difficult to control, and the test results are not very convincing;

[0004] If the recording method is used to switch the scene for multiple tests, the sample data can be improved, but due to multiple manual interventions, it is easy to cause multiple subtle errors, and seriously damage the continuity of the test and lengthen the test cycle;

[0005] In the test of using the server to read the audio running algorithm, the above problems also exist when the audio is recorded in the early stage, and it needs to wait for the audio to be recorded before sending it to the server, and it is impossible to achieve real-time results

Images