Alongside the new version of Loquendo ASR - Automatic Speech Recognition - Loquendo recently released an innovative adaptation tool: the Loquendo Acoustic Model Adaptation. Based on a patented technology, it is now another important element in the Loquendo ASR Tool Suite.

The acoustic models released in any commercial ASR, such as the Loquendo ASR engine, are general models derived from large, telephony-derived corpora, involving many speakers selected to represent a typical distribution in age, sex and geographic area. They represent an average in telephone speech for each language, and guarantee excellent performance in most scenarios.

Where a vocal application is used in a non-standard context, however, acoustic model adaptation can be used to further increase recognition performance.
An application that is always used by the same speaker, for example, can be improved if the acoustic models are adapted to that speaker. In the same way, an application that is used in a particular environment (e.g. in-car) or with a certain kind of input device (e.g. a particular PDA) can be improved if the general models are adapted by the recording of suitable vocal material taken from user interaction.

The most common adaptation scenarios are:

• Speaker
• Audio input channel and input media e.g. specific microphone, PDA
• Environment e.g. in-car, factory
• Speaking style e.g. regional accents, fast speaking
• Application dependent vocabulary e.g. jargon, scientific terms
  

In all cases, adaptation is performed using audio material recorded in the relevant environment/from the relevant source, and can greatly improve recognition performance.

The Acoustic Model Adaptation module released with Loquendo ASR allows acoustic models to be adapted with just a small amount of easily collectable vocal data. In fact, collecting large amounts of speech recordings from specific speakers or in specific conditions, in order to train the acoustic models from scratch, is impractical. Loquendo Acoustic Model Adaptation incorporates a state of the art process for acoustic model adaptation together with an innovative, patented technique.

How Loquendo Acoustic Model Adaptation works