speaker identification
| Software which uses individual characteristics in a speaker's voice (their ‘voice print’) to determine their identity from a set of possible speakers.
| |
| speaker verification (SV)
| As opposed to speaker identification where the system does not know the identity of the speaker, the role of speaker verification is to accept or reject (i.e. consider as an imposter) the supposed identity of the speaker.
| |
|
speaker-dependent (SPKD)
| A service is described as speaker-dependent when it is designed to operate with one speaker who first of all carries out individual training for the system. As opposed to speaker-independent recognition, speaker-dependent recognition uses models which adapt to the voice of an individual user.
| |
| speaker-independent (SPKI)
| A service is described as speaker-independent when it is designed so that anyone who calls the service may use their voice to access it. Speaker-independent recognition uses models which are trained with the voices of a large number of speakers, and do not adapt to individuals.
| |
|
speech traces
| Recordings of all the sound signals (speech and noise) taken into account by the recognition engine during a call to a speech recognition application. Field recordings are used to improve the application. Other studio recordings are used to enhance speech corpora.
Cf. labeling
| |
| substitution
| A recognition error which occurs when one word is recognized as another valid word, for example, if the user says "five" and the system understands "nine".
| |
|
