Adaptive self-trained computer engines with associated databases and methods of use thereof

A method, comprising:

• differentiating between a first speaker and a second speaker in an audio stream of speech by analyzing frequency spectrum characteristics including at least one parameter selected from mean frequency, spectral entropy, spectral centroid, and spectral spread using a multi-core processor;
• converting the audio stream into text by dynamically splitting the audio stream into a plurality of overlapping audio segments with identical non-silent audio portions and processing the overlapping audio segments in parallel through different cores of the multi-core processor at an accelerated transcription rate;
• generating time stamps in the audio stream to associate the text with the audio stream;
• accessing pre-learned phonemes;
• identifying the first speaker based on the pre-learned phonemes and the frequency spectrum characteristics;
• locating a portion of the text associated with the first speaker based on the time stamps;
• segmenting the text associated with the first speaker into text phonemes;
• dynamically comparing phonetically-similar letter pairs of the text phonemes to at least one library of subject-specific terms;
• and
• comparing the text phonemes to the pre-learned phonemes of the first speaker to correct the text associated with the first speaker in real-time based on the pre-learned phonemes by replacing mistakenly transcribed words when a predetermined number of letter pairs match between the mistakenly transcribed word and a subject-specific term, and removing duplicate instances of words in overlapping text transcript segments to generate a cleaned verified transcript.

A non-transitory computer readable medium comprising computer readable instructions that, when executed by a multi-core processor, direct the multi-core processor to:

• differentiate between a first speaker and a second speaker in an audio stream of speech by analyzing frequency spectrum characteristics including at least one parameter selected from mean frequency, spectral entropy, spectral centroid, and spectral spread;
• convert the audio stream into text by dynamically splitting the audio stream into a plurality of overlapping audio segments with identical non-silent audio portions and processing the overlapping audio segments in parallel through different cores of the multi-core processor at an accelerated transcription rate;
• generate time stamps to associate the text with the audio stream;
• access pre-learned phonemes from a subject-specific phonetic pronunciation database;
• identify the first speaker in the audio stream based on the pre-learned phonemes and the frequency spectrum characteristics;
• locate a portion of the text associated with the first speaker based on the time stamps;
• segment the text associated with the first speaker into text phonemes;
• dynamically compare phonetically-similar letter pairs of the text phonemes to at least one library of subject-specific terms;
• and
• compare the text phonemes to the pre-learned phonemes of the first speaker to correct the text associated with the first speaker in real-time based on the pre-learned phonemes by replacing mistakenly transcribed words when a predetermined number of letter pairs match between the mistakenly transcribed word and a subject-specific term, and removing duplicate instances of words in overlapping text transcript segments to generate a cleaned verified transcript.