Learning apparatus, speech recognition rank estimating apparatus, methods thereof, and program
View Patent ↗A learning apparatus comprises a learning part that learns an error correction model by a set of a speech recognition result candidate and a correct text of speech recognition for given audio data, wherein the speech recognition result candidate includes a speech recognition result candidate which is different from the correct text, and the error correction model is a model that receives a word sequence of the speech recognition result candidate as input and outputs an error correction score indicating likelihood of the word sequence of the speech recognition result candidate in consideration of a speech recognition error.
1. A learning apparatus comprising:
processing circuitry configured to
learn an error correction model by a set of a speech recognition result candidate and a correct text of speech recognition for given audio data,
wherein the speech recognition result candidate includes a speech recognition result candidate which is different from the correct text, and the error correction model is a model that receives a word sequence of the speech recognition result candidate as input and outputs an error correction score indicating likelihood of the word sequence of the speech recognition result candidate in consideration of a speech recognition error, and
the processing circuitry learns the error correction model using a set of distributed representations of speech recognition result candidates and word distributed representation sequences of the correct text.
2. The learning apparatus according to claim 1 ,
wherein the set of the speech recognition result candidate and the correct text used for learning the error correction model is composed of a plurality of speech recognition result candidates and one correct text.
3. A non-transitory computer-readable recording medium that records a program for causing a computer to function as the learning apparatus according to claim 1 or 2 .
4. A speech recognition rank estimating apparatus, using an error correction model learned by a learning apparatus, comprising:
processing circuitry configured to
input word distributed representation sequences for word sequences of speech recognition result candidates and distributed representations of the speech recognition result candidates into the error correction model, and finds error correction scores, which are outputs of the error correction model, for respective word sequences of the speech recognition result candidates; and
rank the speech recognition result candidates using the error correction scores,
wherein the learning apparatus learns an error correction model by a set of a speech recognition result candidate and a correct text of speech recognition for given audio data, and the speech recognition result candidate includes a speech recognition result candidate which is different from the correct text, and the error correction model is a model that receives a word sequence of the speech recognition result candidate as input and outputs an error correction score indicating likelihood of the word sequence of the speech recognition result candidate in consideration of a speech recognition error.
5. The speech recognition rank estimating apparatus according to claim 4 ,
wherein the processing circuitry ranks the speech recognition result candidates using scores calculated by weighting and adding speech recognition scores for respective word sequences of the speech recognition result candidates and the error correction scores.
6. A non-transitory computer-readable recording medium that records a program for causing a computer to function as the speech recognition rank estimating apparatus according to claim 5 .
7. A learning method comprising:
learning an error correction model by a set of a speech recognition result candidate and a correct text of speech recognition for given audio data,
wherein the speech recognition result candidate includes a speech recognition result candidate which is different from the correct text, and the error correction model is a model that receives a word sequence of the speech recognition result candidate as input and outputs an error correction score indicating likelihood of the word sequence of the speech recognition result candidate in consideration of a speech recognition error, and
the learning further includes learning the error correction model using a set of distributed representations of speech recognition result candidates and word distributed representation sequences of the correct text.
8. The learning method according to claim 7 ,
wherein the set of the speech recognition result candidate and the correct text used for learning the error correction model is composed of a plurality of speech recognition result candidates and one correct text.
9. A method comprising:
learning an error correction model by a set of a speech recognition result candidate and a correct text of speech recognition for given audio data,
wherein the speech recognition result candidate includes a speech recognition result candidate which is different from the correct text, and the error correction model is a model that receives a word sequence of the speech recognition result candidate as input and outputs an error correction score indicating likelihood of the word sequence of the speech recognition result candidate in consideration of a speech recognition error,
the method further comprising:
inputting word distributed representation sequences for word sequences of speech recognition result candidates and distributed representations of the speech recognition result candidates into the error correction model, and finding error correction scores, which are outputs of the error correction model, for respective word sequences of the speech recognition result candidates; and
ranking the speech recognition result candidates using the error correction scores.