Home > Data Service > Current Project > Speech Data ---Desktop (SDD)

Speech Data ----Desktop (SDD)

The objective of the SDD Program is to develop speech corpora to support speech recognition technology research and development. This project is formed by a series of desktop data collections in over 30 languages, conducted in native countries in a quiet office environment. These corpora were specially designed for the purpose of the training and testing of recent ASR applications and will be licensed upon request.

SDD data overview
Languages: Nearly 30 languages are planned and some new languages will be added based on clients'demands.

NO	Language	Speakers
1	France French	200 people
2	Canadian French	200 people
3	Korean	200 people
4	Japanese	200 people
5	Australian English	200 people
6	UK English	200 people
7	Spain Spanish	200 people
8	Mexico Spanish	200 people
9	USA Mexican Spanish	200 people
10	Italian	200 people
11	German	200 people
12	Arabic	200 people
13	Czech	200 people
14	Danish	200 people
15	Dutch	200 people
16	Finnish	200 people
17	Greek	200 people
18	Hungarian	200 people
19	Norwegian	200 people
20	Polish	200 people
21	Portuguese(European)	200 people
22	Romanian	200 people
23	Swedish	200 people
24	Thai	200 people
25	Turkish	200 people
26	Ukrainian	200 people
27	Vietnamese	200 people

Script Design
All the sentences'scripts were specially designed for speech recognition application training and testing purposes.

Speaker's demographic information
In order to cover as many speaker specific factors as possible in the database recordings, the project is performed in native countries and the following three broad categories are identified for coverage: gender distribution, age distribution and dialectal distribution.

Gender balance
The database will consist of 50% (± 5%) male + 50% (± 5%) female speakers.

Age distribution
For this project, speech data will be collected in the following age categories:

Age group	Minimum # speakers (%)
18 – 30 years	30%
31 – 45 years	20 %
46 – 65 years	15 %

Speakers above 65 or lower than 18 are optional.

    Dialectal regions
The dialectal regions of particular language is carefully identified and a balance between the number of dialectical regions and the number of speakers is specially made to reflect the number of inhabitants of the region for a particular language.

    Recording Conditions
All the speakers were recorded in a quiet office room.

    Data transcribing and annotation
All data was transcribed and annotated based on special rules. Each signal file is accompanied by an ASCII SAM label file which contains the relevant descriptive information.
For the detailed information, please contact us for samples.

Previous：None

Next：None