Keyword confirmation method and apparatus

A method implemented by a computing device having one or more processors and a memory, the method comprising:

• obtaining first audio data via a microphone, the first audio data being recognized as a keyword;
• determining, by the one or more processors and based on silent units pre-stored in a database, a cumulative silence probability of multiple fragments of second audio data contiguous in time with the first audio data, including;
• determining a pronunciation similarity probability of individual fragment of the multiple fragments based on pronunciation units,
• determining a silence probability for individual fragment, the silence probability being a similarity probability between the pronunciation units and the silence units corresponding to a largest similarity probability, and
• determining the cumulative silence probability of the multiple fragments based on the silence probability for the individual fragment;
• determining, by the one or more processors, a cumulative keyword probability of multiple fragments of the first audio data;
• and
• confirming, by the one or more processors, the first audio data as an effective keyword upon determining that a relationship between the cumulative silence probability and the cumulative keyword probability satisfies a second preset condition,
• wherein the silent units are obtained by iteratively training a model using training data.

An apparatus comprising:

• one or more processors;
• memory;
• an audio data obtaining module stored in the memory and executable by the one or more processors configured to cause the one or more processors to obtain first audio data, the first audio data being recognized as a keyword;
• a silence fragment determining module stored in the memory and executable by the one or more processors configured to cause the one or more processors to determine that multiple contiguous silence fragments exist in second audio data contiguous in time with the first audio data, including:
• partitioning the second audio data into a plurality of fragments,
• determining a pronunciation similarity probability of individual fragment,
• the pronunciation similarity probability being a largest similarity probability among pronunciation units,
• determining a silence probability for individual fragment, the silence probability being a similarity probability between the pronunciation units and the silence units corresponding to a largest similarity probability, and
• determining that the individual fragment is a silence fragment upon determining that a relationship between the pronunciation similarity probability and the silence probability of the individual fragment satisfies a preset condition;
• and
• an effective keyword determining module stored in the memory and executable by the one or more processors configured to cause the one or more processors to confirm the first audio data as an effective keyword,
• wherein the silent units are obtained by iteratively training a model using training data.