System and method for detection and correction of incorrectly pronounced words
A system and method are disclosed for capturing a segment of speech audio, performing phoneme recognition on the segment of speech audio to produce a segmented phoneme sequence, comparing the segmented phoneme sequence to stored phoneme sequences that represent incorrect pronunciations of words to determine if there is a match, and identifying an incorrect pronunciation for a word in the segment of speech audio. The system builds a library based on the data collected for the incorrect pronunciations.
1. A method of recognizing incorrect pronunciations of words in speech, the method comprising:
capturing a segment of speech audio;
performing phoneme recognition on the segment of speech audio, according to a phonetic model, to produce a digital representation of a phoneme sequence;
segmenting the digital representation of the phoneme sequence into a word sequence according to a pronunciation dictionary having words pronunciations tagged as incorrect pronunciations;
producing an output word sequence with a corresponding indication that at least one word in the output word sequence had an incorrect pronunciation in the captured segment of speech audio; and
updating the pronunciations in the pronunciation dictionary using training data derived from the word sequence with the incorrect pronunciation.
2. The method of claim 1 further comprising storing the incorrect pronunciation of the word in a log of words spoken with incorrect pronunciation.
3. The method of claim 2 further comprising performing a statistical classification on the log, when the log is specific to a speaker, to compute a profile property for the speaker.
4. The method of claim 2 , wherein the stored incorrect pronunciation of the word is specific to a speaker.
5. The method of claim 4 , wherein the word is represented as characters based on an abugida writing system.
6. The method of claim 4 , wherein the word is represented as characters based on a logographic writing system.
7. The method of claim 1 further comprising the steps of:
correcting the segment of speech audio by replacing the incorrect pronunciation of the word with a correct pronunciation to produce a corrected word sequence; and
synthesizing, using the corrected word sequence, a synthesized segment of speech audio.
8. A system for training speakers of a second language on correct pronunciations, the system comprising:
an input device that receives, from a user, a segment of speech audio that includes a plurality of words, at least one word having been spoken with an incorrect pronunciation;
a phoneme-based automatic speech recognition (ASR) component that performs speech recognition on the segment of speech audio to produce a word sequence and a corresponding indication that a word from the plurality of words is incorrectly pronounced by the user; and
a speech synthesis component to produce a spoken audio response that includes a correct pronunciation of the incorrectly pronounced word,
wherein the spoken audio response helps the user learn the pronunciation of the word by repeating the incorrectly pronounced word with a correct pronunciation and the spoken audio response provides a syllable stress that differs from the incorrectly pronounced word having an incorrect syllable stress from a correct pronunciation of the incorrectly pronounced word.
9. The system of claim 8 further comprising a storage component for storing an indication of the incorrectly pronounced word in a log of words spoken with incorrect pronunciations such that pronunciation errors can be analyzed.
10. The system of claim 8 further comprising means to recognize an incorrect pronunciation by detecting incorrect syllable stress of the word.
11. The system of claim 8 further comprising a display component for displaying a transcription of words from the request, wherein the displayed transcription includes the mispronounced word in context with respect to other words with emphasis on the incorrectly pronounced word.
12. The system of claim 8 further comprising a display component for displaying the incorrectly pronounced word with a correct pronunciation for the incorrectly pronounced word shown using written characters from the user's native written language.
13. The system of claim 12 , wherein the written characters are part of a logographic writing system.
14. The system of claim 12 , wherein the written characters are part of an abjad writing system.
15. The system of claim 12 , wherein the written characters are part of an abugida writing system.