Speech recognition apparatus, control method, and non-transitory storage medium

A control method performed by a computer and comprising:

• inputting an acquired audio frame to a first model converting an audio frame into compressed character string data to thereby generate the compressed character string data corresponding to the acquired audio frame;
• and
• inputting the compressed character string data to a second model converting the compressed character string data into character string data to thereby generate the character string data corresponding to the acquired audio frame, wherein
• a data size of the compressed character string data is smaller than a data size of the character string data,
• the first model is a model learned by training data with an audio frame as input data, and with, as correct answer data, compressed character string data acquired by encoding character string data represented by the audio frame,
• the second model is a learned decoder acquired by learning an autoencoder being constituted of an encoder converting input character string data into compressed character string data, the decoder converting, into character string data, compressed character string data output from the encoder,
• a character vector constituting the character string data is a N-bit vector, and
• a character vector constituting the compressed character string data is a vector having a size smaller than an N-bit.

A speech recognition apparatus comprising:

• at least one memory storing instructions;
• and
• at least one processor configured to execute the instructions to perform operations comprising:
• inputting an acquired audio frame to a first model converting an audio frame into compressed character string data to thereby generate the compressed character string data corresponding to the acquired audio frame;
• and
• inputting the compressed character string data to a second model converting the compressed character string data into character string data to thereby generate the character string data corresponding to the acquired audio frame, wherein
• a data size of the compressed character string data is smaller than a data size of the character string data,
• the first model is a model learned by training data with an audio frame as input data, and with, as correct answer data, compressed character string data acquired by encoding character string data represented by the audio frame,
• the second model is a learned decoder acquired by learning an autoencoder being constituted of an encoder converting input character string data into compressed character string data, the decoder converting, into character string data, compressed character string data output from the encoder, and
• a number of nodes of an input layer in the decoder is smaller than a number of nodes of an output layer in the decoder.