Method, apparatus, electronic device, and medium for speech processing

An electronic device comprising:

• at least one processing unit;
• at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit which, the instructions, when executed by the at least one processing unit, cause the device to perform acts comprising:
• generating a token-level semantic feature sequence of target speech data based on a frame-level acoustic feature of the target speech data, a semantic feature of the token-level semantic feature sequence corresponding to a token of the target speech data;
• generating a token-level voiceprint feature sequence of the target speech data based on the frame-level acoustic feature, a voiceprint feature of the token-level voiceprint feature sequence corresponding to a token of the target speech data, comprising:
• generating the token-level voiceprint feature based on a frame-level voiceprint encoded feature and a set of weights, comprising:
• partitioning the frame-level voiceprint encoded feature based on a comparison of an accumulated value of successive weights in the set of weights with a threshold;
• and
• generating a token-level voiceprint encoded feature based on the partitioned frame-level voiceprint encoded feature and the set of weights;
• obtaining the token-level voiceprint feature by performing voiceprint decoding of the token-level voiceprint encoded feature;
• and
• determining a token in the target speech data where a speaker change occurs based on the token-level semantic feature sequence and the token-level voiceprint feature sequence.

A computer program product having a non-transitory computer readable medium which stores a set of instructions to perform data access;

• the set of instructions, when carried out by computerized circuitry, causing the computerized circuitry to perform a method of:
• generating a token-level semantic feature sequence of target speech data based on a frame-level acoustic feature of the target speech data, a semantic feature of the token-level semantic feature sequence corresponding to a token of the target speech data;
• generating a token-level voiceprint feature sequence of the target speech data based on the frame-level acoustic feature, a voiceprint feature of the token-level voiceprint feature sequence corresponding to a token of the target speech data, comprising:
• generating the token-level voiceprint feature based on a frame-level voiceprint encoded feature and a set of weights, comprising:
• partitioning the frame-level voiceprint encoded feature based on a comparison of an accumulated value of successive weights in the set of weights with a threshold;
• and
• generating a token-level voiceprint encoded feature based on the partitioned frame-level voiceprint encoded feature and the set of weights;
• obtaining the token-level voiceprint feature by performing voiceprint decoding of the token-level voiceprint encoded feature;
• and
• determining a token in the target speech data where a speaker change occurs based on the token-level semantic feature sequence and the token-level voiceprint feature sequence.