Symbol sequence estimation in speech

An apparatus comprising:

• a processor;
• and
• one or more computer readable mediums collectively including instructions that, when executed by the processor, cause the processor to perform operations comprising:
• applying two or more templates for extracting a predetermined number of symbol sequences from speech-to-text data;
• estimating, by employing a trained estimation model, whether each candidate of a plurality of candidates from the speech-to-text data is a target symbol sequence;
• and
• using a corresponding related portion including a repetition indication of each of the plurality of candidates, wherein each candidate is divided into a plurality of word groups, with each of the plurality of word groups being fed into one of a plurality of Long Short-Term Memory (LSTM) neural networks in a direction of an inverse order of the text to generate a plurality of outputs to estimate the target symbol sequence from the speech-to-text data with the trained estimation model using less computational resources.

A non-transitory computer readable storage medium having instructions embodied therewith, the instructions executable by a processor or programmable circuitry to cause the processor or programmable circuitry to perform operations comprising:

• applying two or more templates for extracting a predetermined number of symbol sequences from speech-to-text data;
• estimating, by employing a trained estimation model, whether each candidate of a plurality of candidates from the speech-to-text data is a target symbol sequence;
• and
• using a corresponding related portion including a repetition indication of each of the plurality of candidates, wherein each candidate is divided into a plurality of word groups, with each of the plurality of word groups being fed into one of a plurality of Long Short-Term Memory (LSTM) neural networks in a direction of an inverse order of the text to generate a plurality of outputs to estimate the target symbol sequence from the speech-to-text data with the trained estimation model using less computational resources.