The Grammar Tester compares transcripts of interactions to the output of a speech engine, and returns an alignment type. The alignment type gives information on how well the speech engine recognized the spoken input. The alignment types are:
MATCH (m): The speech engine recognize the same words as the transcript.
|
Transcript: |
TWO |
THREE |
|
ASR Result: |
TWO |
THREE |
|
Alignments: |
(m) |
(m) |
: The speech engine recognized more words than the transcript.
|
Transcript: |
* |
TWO |
THREE |
|
ASR Result: |
ONE |
TWO |
THREE |
|
Alignments: |
(i) |
(m) |
(m) |
DELETION (d): The speech engine recognized fewer words than the transcript.
|
Transcript: |
ONE |
TWO |
THREE |
|
ASR Result: |
* |
TWO |
THREE |
|
Alignments: |
(d) |
(m) |
(m) |
SUBSTITUTION (s): The speech engine recognizes a word different from the transcript
|
Transcript: |
TWO |
THREE |
|
ASR Result: |
FOUR |
THREE |
|
Alignments: |
(s) |
(m) |
OOV_INSERTION (OI): The transcript's word is out-of-vocabulary but the engine adds an extra word.
In this example, "know" is not in the grammar so it's considered to be out-of-vocabulary.
|
Transcript: |
KNOW |
|
ASR Result: |
DELTA |
|
Alignments: |
(oi) |