AudioMandarin Chinese

Conference Data

About this dataset

1. Product Overview

The conference speech audio database contains 1,459 hours of recordings sourced from online conferences. These are authentic recordings captured at the central control console of actual meeting venues, with an audio sampling rate of 16 kHz and 16-bit resolution. The transcribed texts cover 16 domains, including technology, finance, education, and more. The data has been professionally transcribed and annotated by skilled linguists and rigorously quality-checked, achieving a text accuracy rate exceeding 97%.

2. Audio Corpus

Corpus Construction

Corpus Content

To ensure diversity, the corpus was curated across 17 distinct fields.

Serial Number · Domain ID · Content

1 · MTGC · Technology

2 · MTGA · Finance

3 · MTGD · Healthcare

4 · MTGE · Sports

5 · MTGF · Education

6 · MTGG · Entertainment

7 · MTGI · Real Estate

8 · MTGH · Literature & Arts

9 · MTGJ · Transportation & Logistics

10 · MTGM · Government Reports

11 · MTGK · Automotive

12 · MTGB · Lifestyle Services

13 · MTGO · Entrepreneurship Management

14 · MTGL · E‑commerce Retail

15 · MTGN · Interviews & Speeches

16 · MTGP · Multimedia

17 · MTGQ · Other

Corpus Processing

De-identification: Removing politically sensitive, personally identifiable, or sexually explicit content.
Removal of symbols such as <, >, [, ], ~, /, \, =.
Exclusion of materials containing languages other than Chinese and English.
Elimination of single sentences exceeding 25 characters in length.
Standardization of formatting.

Text Structure Design

To ensure comprehensive phonetic coverage and balanced phoneme distribution, the corpus encompasses 17 distinct domains, structured as follows:

Serial Number · Domain · Allocation per Domain (%)

1 · Technology · 43.78

2 · Finance · 9.54

3 · Healthcare · 3.28

4 · Sports · 0.72

5 · Education · 3.94

6 · Entertainment · 2.55

7 · Real Estate · 0.54

8 · Literature & Arts · 1.49

9 · Transportation & Logistics · 0.24

10 · Government Reports · 1.25

11 · Automotive · 1.59

12 · Lifestyle Services · 2.41

13 · Entrepreneurship Management · 3.31

14 · E‑commerce Retail · 1.78

15 · Interviews & Speeches · 8.81

16 · Multimedia · 1.97

17 · Other · 12.8

Total · · 100

3. Speech Data Transcription

Transcribers convert audio into written text, striving to maintain complete alignment between spoken words and their corresponding transcription. General guidelines include: 1) Transcriptions must exactly match the audio; no extra words, omissions, or errors are permitted. 2) Numerals should be spelled out in Chinese characters—for example, “one two three” rather than “123.” Distinguish carefully between “one” and “yao,” and between “two” and “liang.” 3) Any English pronunciation within the audio should be rendered either in Chinese characters or left as English, depending on context. Specifically: All letters or words appearing in URLs must be capitalized—for instance, “www.abc.com” becomes “Three W dot A B C dot com.” English words embedded in the audio should be transcribed entirely in lowercase. Individual English letters should be capitalized. Certain proper nouns or acronyms requiring all capitalization should retain spaces—such as “C E O,” “C C T V,” etc. 4) Annotations must fully reflect the original pronunciation without any deletions.

4. Data File Directory

Directory Structure

The hierarchical structure of the data directory is as follows:

└─DOC (Documentation Files)
││├─all_wav_list.txt (Audio File List)
││├─content.txt (Transcription List)
│
└─SPEECHDATA (Data Folder)
│├─MTGA (Domain Folder)
││ ││ MTGA0036A001W00131.wav (Audio File)