Speaking practice system with reliable pronunciation evaluation

A speaking practice system with reliable pronunciation evaluation, including:

• a data preprocessing system and a goodness of pronunciation evaluation system running on a data processing server,
• wherein:
• the data preprocessing system includes a phonetic symbol generation system and an audio recording preprocessing system, wherein, according to a practice text as well as an audio recording of a user reading the practice text, the phonetic symbol generation system generates phonetic symbols, and the audio recording preprocessing system generates audio data;
• the goodness of pronunciation evaluation system scores the audio recording based on the phonetic symbols and the audio data;
• the phonetic symbol generation system operates an artificial intelligence model, which is configured to generates the phonetic symbols in response to the practice text including at least one of polyphonic words, wherein the polyphonic words are words having at least two pronunciations and each one of the at least two pronunciations corresponds to a part of speech or a special usage;
• the artificial intelligence model includes a plurality of audio-symbol models corresponding to different polyphonic words;
• and
• in response to the practice text being an individual word belonging to the polyphonic words, the corresponding audio-symbol model, configured to receive the audio recording and output the phonetic symbols, operates to generate the phonetic symbols according to confidence values of different pronunciations of the individual word that the audio-symbol model evaluates from the audio recording.