Speech Processing

Drucken PDF

Speech Processing Services

Our Professional Services team offers for German, French, English and Arabic the following services:

Design of speech databases
We set up recording protocols, create the appropriate reading scripts and plan the most suitable speaker recruitment strategy according to the aim of your application.

Recording of speech databases
With our know how in projects like Orientel we record dialogues, over a telephone as well as a microphone, in different environments.

Annotation of speech databases
Our long years experience allow us to develop methods and procedures providing efficient and consistent annotations. We can provide orthographic or phonetic transcriptions, transcriptions of non-speech events, part-of-speech tags, alignment of speech signals and transcriptions, depending on your system. Our high qualified stuff of Arabic native speakers allows us to deliver high quality.

Text-to-Speech Research

Our goal is to define a coarse transcription that can be performed quickly, but covers adequately the acoustic events most important for the training and testing of automatic speech recognisers. The transcription is fully orthographic (vocalized) and includes a few details that represent audible acoustic events (speech and non-speech) present in the corresponding waveform files. The symbol set aims at the localisation of the main acoustic events according to a coarse categorisation rather than a full description of all possible sounds that may appear during a recording. Extra marks contained in the transcription aid in interpreting the text form of the utterance; markers for non-speech acoustic events and distortions have been chosen such that they can be automatically removed or modified to yield the base transcription. The overall aim is to keep as much speech in the corpus as possible and to avoid the need for deleting recordings from the corpus due to some extra noises, disfluencies, etc.

Validation of speech databases
We check the quality of collected databases by different partners in order to assure high level of quality. We provide a quality stamp for resulting databases in a systematic manner. We provide advice on how to obtain a high quality level of databases.

Creation of phonetic lexicons
We have high qualified expertise in creating phonetic lexicons for Arabic language as well as special names lexica.

Speech recognition

  • Speech recognition Application specification in close collaboration with the customer.
  • Dialogue design covering menus, vocabularies, grammars and voice prompts
  • Development of applications combined with Text-To-Speech.
  • User Interface design for Arabic speech applications.
  • Computer telephony integration.
  • Dictation.