TA^Ì Powered by SpeechMagic Glossary

Acoustic Adaptation/ARF file

Adaptation process which continuously improves an author’s acoustic references.  This is done by analyzing dictations and automatically updating the ARF (Acoustic Reference File) to understand each author’s voice better.

Acoustic References

Statistical data which describes the voice characteristics of an individual user.  This can include the user’s accent, pronunciation, input device, etc.  Speech recognition uses this data to interpret an author’s speech input; it is stored in an ARF.

ASR (Automatic Speech Recognition)

        Automatic conversion of speech to text via a software program.

Audio Wizard

Tool for adjusting the settings of your audio system to ensure the best possible sound quality for speech recognition.  These include setting the recording volume, playback volume, and the VA (voice activated recording) level.

Background Lexicon

Dictionary containing between 300,000 and 800,000 words, whose usage is not considered frequent enough for inclusion in a specific ConText Lexicon.  Used for reference when unknown words are added to the ConText during ConText Adaptation.

Background Recognition

Speech recognition which takes place during the dictation but where the recognized text is not displayed to the author onscreen.  Can also be referred to as ‘hidden online’ or ‘backend’ recognition.

Backend

Any process which takes place on a server rather than on a client PC (e.g. recognition, adaptation, etc.). Can also be referred to as ‘central’ or ‘offline’.  Opposite of frontend.

Command Grammar

Collection of words or phrases which are not recognized as text, but upon recognition TA^Ì integration executes a specific action.  For example, text formatting commands and document navigation commands will be grouped into their own respective command grammars.

ConText or ConText Lexicon

Speech recognition systems product which contains the vocabulary (ConText Lexicon) and the default language model for speech recognition.  It contains words and word combinations, as well as information on pronunciation.  ConTexts are specific to one specialty and one language.

ConText Adaptation Task

System task which analyzes corrected dictations for words and grammar items not contained in the ConText, adds unknown words to the ConText and updates the user’s language model in order to improve the recognition rate.  It can use dictations which have been recognized and corrected, or documents not created using speech recognition (e.g. imported via the ConText Tuner).

ConText Tuner

ConText Adaptation Function which enables the import and analysis of documents or files for words or phrases not contained in a ConText.

Correction

Editing a document created by speech recognition in order to replace incorrectly recognized words or phrases.  Correction may be performed by a designated correctionist or the author, and can be aided by functions such as synchronous playback and recognition alternatives.

Correctionist

A medical language specialist, aka transcriptionist, who has been trained to edit transcripts created through speech recognition. 

Frontend

Process which takes place on the client rather than the server, (e.g. frontend recognition, adaptation, etc).  Opposite of backend.

General Correspondence ConText

A ConText which contains vocabulary for general dictation purposes, such as letters and memos. 

Grammar Item

Word or phrase which the recognizer knows to treat differently, with reference to a ConText grammar rather than the Language Model.  For example, numbers which represent a date or a dosage are typical grammar items.

Initial Language Model

      Contains information on the combinations of words that are most likely to occur.

Initial Training

Process during which an author-specific ARF (Acoustic Reference File) is created from training texts read aloud.  This can take from as little as two minutes to over an hour, depending on the user.

Online Recognition or frontend

Recognition of speech to text which takes place during the dictation, and where the recognized text is displayed simultaneously to the author.  Opposite of batch recognition.  Can also be referred to as ‘frontend’ or ‘simultaneous’ recognition.

Normal

            Pre-defined text often used in radiology transcription to insert an author’s typical language on a particular type of test.   

Professional ConText

ConText for specific professional fields, e.g. medical, legal insurance.  Compare: General Correspondence ConText.

Recognition Rate

Percentage of words correctly recognized when dictating.  Sometimes referred to as recognition accuracy.

Recognition Result

Text generated by the recognition of dictated speech, including ancillary data such as sound file position and recognition alternatives.

Redundant Phrase

Word or re-occurring words that are recorded but should not appear the final report, e.g. thank you, etc.

Sound File

File which contains audio data in a special format, e.g. .wav or .dss

SpeechMike or Handmike

Industry standard audio device designed and built by Philips specifically for dictation and speech recognition.  This device includes a microphone and speaker and can also incorporate a trackball or barcode scanner.

SRD file (Speech Recognition Document)

File format for storing speech recognition dictations. An SRD file usually contains sound, text and additional information, including the author’s name, the name of the ConText used for speech recognition, the recognized and corrected text, etc.

TA^Ì Powered by SpeechMagic Systems Tasks

Service which performs a task in the TA^Ì Powered by SpeechMagic system.  System tasks include: Recognition Task, ConText Adaptation Task, Purge Task.

User Language Model

This is the same as the ILM, initial language model, but is specific to a user.  As the system learns, this file will grow with word combinations that are specific to the user.

User Lexicon

      Words and their phonetic transcriptions.

View and Print as PDF of This Page