Bavarian Archive for Speech Signals - Services Schiel

[Corpora Production]
Corpora Validation
Pronunciation Dictionaries
System Evaluation
Seminars & Lextures

Speech Corpora Production

BAS Services Schiel (BASSS) has a long experience in the efficient production of technical speech corpora.
What is a speech corpus?

A speech corpus is a collection of controlled speech recordings together with their respective annotations and meta information. There exists a wide range of possible parameters when specifying a new speech corpus, such as: technical format and quality, surrounding acoustics, linguistic/situational context, speaker profiles, content, background noise, purpose of the corpus (ASR, synthesis etc.).

Speech corpora are needed for a number of empirical methods in speech processing such as: Automatic Speech Recognition, Speaker Verification/Recognition, Concatenative Speech Synthesis. Furthermore, companies that are planing to include such technologies into their products are in need of specific test corpora to evaluate their developed/acquired technology.

During the last decade the BAS has produced most of the existing German speech corpora (see the BAS online catalogue for details). The founders of BASSS have compiled these experiences into a recommendation for the production of speech corpora which is used as a guideline for our ongoing and future corpora productions.

What can BASSS do for you?

BASSS offers a full support on all stages of the speech corpora production process: Specification, Preparation, Speaker Recruitment, Pre-Validation, Recording, Post-processing, Annotation, Distribution, Validation.

In some cases our customers choose to have the full control over parts of the process (e.g. for legal reasons), while other parts are contracted to BASSS.
For example, the specification and the speaker recruitment are provided by BASSS, the actual recordings are performend by the customer (because of insurance issues) while the orthographic transliteration, the segmentation and labeling and the postprocessing are again done by BASSS

Since BASSS acts as a private company and is not represented by the university, contracted legal restrictions and non-disclosure agreements about the contents of the speech corpora are possible.

Speaker Database

BASSS maintains a large speaker database of all speakers that have ever participated in a BAS corpora production. This enables us to recruit new speakers groups very effectively. Our response rate to calls to our registered speakers is about 34% (which is a very high value compared to the response rate of 1% by unsolicitated mail).

Recording and Annotation Techniques
BASSS has developed a number of widely used recording and annotation techniques, file formats and quality assurance methods that enable us to minimize logistic efforts and errors during a speech corpora production. Please refer to our scientific publications for a closer look at WebTranscribe, MAUS and SpeechRecorder.

Copyright 2005 BAS Services Schiel

Impressum: Florian Schiel, Moltkestr. 1, D-80803 München, Germany,