Speech Dataset Collection and Annotation

We provide a worldwide collection of speech datasets that are diverse, scalable, and meticulously transcribed, perfect for training machines to accurately recognize and understand different types of languages.

Speech Database

French-Algeria Speech Datasets

Reading

330 Speakers

211 Hours

Arabic-Saudi Arabic Speech Dataset-2

Reading

131 Speakers

82 Hours

English-US Speech Dataset

Reading

1935 Speakers

865 Hours

Chinese-Mandarin-English Speech Dataset Co-Switch

Reading

8477 Participants

4089 Hours

English-US Call Center Speech Dataset

Scene: Live

Age: >16 years old

287 Hours

Chinese-Mandarin-LiveStream Speech Datasets

Natural Language

5079 Hours

Scene: Live

Mandarin-China Children Speech Dataset

Mandarin-China

1,105 Hours

10,060 Speaker Number

How it Works

Algorithm

Development

Data Demand

Generation

Dataset

Definition/Design

Trial and

Improvement

Mass

production

Quality Control

Data Package

Delivery

Data Collection and Annotation

We have all kinds of audio classification dataset,for example speech command dataset, common voice dataset.Beside this there is North American voice dataset,African sound dataset，Asian audio datasets，European voices dataset.

Environments

Indoor

Studio

Outdoor

Incar

Devices

Mobile (iOS/Android)

Computer (Desktop/Laptop)

Pro (Hi-Fi recorder/Mic Array)

Speakers

Language: Chinese/English/French/German…

Gender balanced: 1:1

Age: Children/Senior

Education Background

Machine

annotate

Human

transcribe / Validate

Rounds QA by

human & machine

Voice Dataset Annotation

Accuracy between 95%~98%

Surfing Tech applies its own algorithm during speech dataset annotation to ensure high efficiency and accuracy. We achieve above 95% accuracy rate after three rounds of quality inspection, which makes the audio datasets more valuable for speech emotion recognition dataset, semantic understanding, and human-computer interaction.

Speech Data Portfolio