Datasets:
extra_gated_prompt: Please fill out the following information to verify your identity.
extra_gated_fields:
Name: text
Phone Number: text
Email: text
Organization or School: text
Major: text
Education level: text
Identity:
type: select
options:
- Student
- Teacher
- Employee
extra_gated_heading: Acknowledge license to accept the repository
extra_gated_description: >-
Please fill in the information truthfully and keep your phone line clear.
After verifying your information, you will be able to use our dataset.
extra_gated_button_content: Acknowledge license
task_categories:
- voice-activity-detection
- automatic-speech-recognition
- audio-to-audio
language:
- zh
- en
license: cc-by-nc-nd-4.0
Dataset Introduction
Cantonese dialect conversation
Version
v1.0
Release Date
2024-11-04
Data Description
Material Type Voice:Collection and Annotation
Data Description (including: language category, etc.):Cantonese colloquial speech
Total Data Volume:655 hours
Collection Equipment Requirements (model, etc.):Telephone recording
Collection Environment Requirements:Indoor
Pronunciation Style (Neutral news, emotional, reading/Free topic, natural speech):Natural speech
Number of Collectors:228 people
Recording Content Audio scenes include: Customer service dialogue scenarios, financial industry, simulated production scenario data
Duration of Each Recording:5-10 minutes
Annotation Content:Text, timestamp, gender, background noise, English, amplitude reduction
Annotation Format:Customer service and customer are annotated separately
Recording Format (whether to follow reading, whether to read aloud, whether dialogue, etc.; Follow reading is only for children):Dialogue
Voice Ability Requirements (requirements for accent; requirements for regional distribution; if it is a small language, its language ability requirements; whether professional broadcasters are needed):Normal Cantonese colloquial speech
Recording Environment (Recording studio/Signal-to-Noise Ratio SNR[Office/Shopping mall/Outdoor/Desk/Mobile phone/Car (model, speed, in-car environment)]):Office
Audio Format: 8k, 16bit wav
Channels: Dual channel, left channel, right channel
Silent Reservation:None reserved
Voice Annotation:Divided according to the speaker's pauses and marked with corresponding text
Acceptance Rate Requirement:99%
## Directory Structure
root_directory/
├── audio/
│ ├── audio1.wav │ ├── text1.txt