Very low accuracy

[zyc1310517843]

Hello, I downloaded the official Chinese speech model. It seems that the recognition rate is very low and the basic recognition is not correct. Thank you for your guidance.

[lvan-jone]

Where did you download the official Chinese package? If the recognition rate is low, you need to get tools like dictionary models yourself

[SwimmingTiger]

I think he downloaded this model: https://sourceforge.net/projects/cmusphinx/files/Acoustic%20and%20Language%20Models/Mandarin/

I am also using this model. The dictation accuracy of this model is indeed very poor. In dictation mode, it can hardly generate any readable sentences. I can only get some unrelated fragments of words.

But if a JSGF grammar file is loaded, the accuracy is acceptable. Note: It seems that manual word segmentation is required for Chinese grammar file, that is, adding spaces between each word in the sentence. Otherwise, The dictionary is missing a phonetic transcription for the word 'xxxxxxxxxxxxxxxxxx' will be reported and you will not be able to identify any content.

[SwimmingTiger]

Or we can make some adjustments to the acoustic model or configuration to improve the accuracy of dictation.

• Adapting the default acoustic model: https://cmusphinx.github.io/wiki/tutorialadapt/
• Tuning speech recognition accuracy: https://cmusphinx.github.io/wiki/tutorialtuning/