Speak&Find Technical Specifications

General Features:

Open Database Interface: Allows integration into any DB environment (ODBC, DB2, Oracle, SQL Server, ...). If you have specific questions regarding your database integration please contact our consulting department.

C/C++ API: Easy to use and fast interface to optimize performance and portability.

Online Help: Comprehensive guide and reference manual.

Sample applications: Sample application to quickly understand all functionalities and how to integrate them into your application.

Portability: The design and implementation Speak&Find is specially designed for easy portability. Please contact our development department for further information.

Scalability: Speak&Find can be integrated easily in large scale systems with multiple servers as well as in small devices like tablet PCs or PDAs.

Philips speech recognition engine: Philips' HSA IV technology, latest transcription technology.

Online Help: Comprehensive guide and reference manual.

Compure audio mining technology: Compure's PHONS technology provides highest accuracy for indexing and searching in large sound databases.

Supported Languages:

US English, UK English, 

German (special Acoustic Models for Austrian German are available), 

French, 

Dutch, Flemish, 

Spanish (special Acoustic Models for Catalan, North-American and South-American Spanish are available),

Portuguese (special Acoustic Models for Brazilian and European Portuguese are available), 

Swedish, 

Italian, 

Finnish

Supported Audio Formats:

PCM 8kHz 16bit

PCM 16kHz 16bit

Average Accuracy:

The following accuracies were measured for a speaker independent system with telephone input, PCM 8kHz 16Bit:

Attention: All accuracy values depend on audio quality, speech quality, words / phrases to be searched for.

 

Service	Accuracy
Indexing/Searching/Monitoring	99.4%
Quick Search	90%(*)
Transcription	95%(*)

(*) The accuracy for Quick Search and Transcription service are dependent on the available ConTexts (Vocabulary) used for the test. For transcribing any general purpose text accuracy levels of 30%-50% are feasible. Therefore Compure recommends to apply or develop specific ConText to match the requested domain.

Average Speed:

On a PC with a Pentium IV 2 GHz processor, 512MB memory, running Windows 2000 the following Real Time Factors(*) were measured:

 

Service	Real-time
Indexing/Searching/Monitoring	7.3
Quick Search (*)	82.2
Transcription (*)	2.3

(*) Real-Time Factor 1 means: For processing 1 hour of stored sound the service takes 1 hour. A Real Time Factor 3 means: For processing 1 hour sound the service takes 20 min (3 times as fast).

Required Hardware and Operating Systems:

Speak&Find is a pure software product and can be used on any IBM compatible PC. The following operating systems are supported:

• Windows 98

• Windows NT

• Windows 2000

• Windows XP