AudioMandarin Chinese

Mandarin Chinese Speech Database

About this dataset

1. Product Overview

The Chinese Mandarin speech database contains 1,000 hours of audio, with 718 hours from [ZH-CN] and 282 hours from [ZH-CN]. The recorded transcripts cover 12 domains, including wake-up words, voice control commands, smart home applications, autonomous driving, industrial production, and more. Recordings were conducted in a quiet indoor environment using three different devices: a high-fidelity microphone (44.1 kHz, 16-bit), an Android smartphone (16 kHz, 16-bit), and an iOS smartphone (16 kHz, 16-bit). Voice data was primarily captured using iOS smartphones. A total of 1,991 speakers from various accent regions across China participated in the recording process. After transcription and annotation by professional linguists, and rigorous quality assurance checks, the database achieves a text accuracy rate exceeding 96%.

2. Recording Corpus

Corpus Construction

Corpus Content

Considering the application of speech recognition in areas such as smart homes, autonomous driving, and industrial production, the corpus was curated across 12 distinct domains.

Serial No. · Domain

1 · Home Appliances Names and Control Commands

2 · POI (Geographic Information)

3 · Music and Radio (Voice-Controlled)

4 · Digital Products (Voice-Controlled)

5 · TV and Movie Titles

6 · Finance

7 · Technology

8 · Sports

9 · Entertainment

10 · Current Affairs News

11 · English Spelling

12 · General Text

Corpus Processing

De-identification: Removal of politically sensitive, personally identifiable, or pornographic/violent content.
Removal of symbols such as <, >, [ , ], , /, \, =.
Exclusion of materials containing languages other than Chinese and English.
Standardization of formatting.

3. Speaker Information

Basic Information Records

Speaker information includes task ID, gender, accent region, age range, and place of origin.

Task ID · Age Range · Gender · Accent Region

C0002 · B · Male · North

Each speaker is assigned one unique task ID, corresponding to a single transcript. Each speaker may participate only once. Accent regions are categorized into Northern, Southern, and Other based on the speaker’s native language area. Age ranges are defined as follows: A (under 16), B (16–25), C (26–40), D (41+).

Speaker Demographic Characteristics

Gender Ratio

The database comprises 1,991 speakers—845 males and 1,146 females.

Gender · Male · Female · Total

Ratio · 42.4% · 57.6% · 100%

Age Distribution

Ages 16–25: 316 speakers; Ages 26–40: 71 speakers; Over 41: 13 speakers.

· Age Group · Number · Proportion · Male · Female

A · Under 16 · 16 · 0.8% · 5 · 11

B · 16–25 · 1707 · 85.7% · 722 · 985

C · 26–40 · 201 · 10.1% · 89 · 112

D · Over 41 · 67 · 3.4% · 29 · 38

Total · 1991 · 100% · 845 · 1146

Dialect Distribution

Northern dialect: 1,247 speakers; Southern dialect: 679 speakers; Other dialects: 65 speakers.

Region · Number · Proportion

Northern · 1275 · 64%

Southern · 687 · 34.5%

Other · 29 · 1.5%

Total · 1991 · 100%

4. Data Recording Environment

Recording Environment

The recording took place in a quiet indoor setting, replicating real-life home scenarios without significant background voices or other noise, and free from echoes. Speakers read aloud their scripts at normal speaking speed.

Recording Equipment

Equipment included a high-fidelity microphone and recorder, an iOS smartphone, and an Android smartphone. All recordings were captured using iOS devices.

Recording Method

Speakers stood approximately 20 cm away from the high-fidelity microphone, speaking at normal volume and pace while reading their scripts. Android and iOS smartphones were positioned 20 cm away from the microphone respectively.

5. Speech Data Transcription

Transcribers converted audio into written text, striving for complete alignment between spoken content and written form. General guidelines include: 1) Transcriptions must exactly match the spoken audio—no extra words, omissions, or errors. 2) Numerals should be transcribed in Chinese characters, e.g., “一二三” instead of “123.” Pay attention to distinguishing “一” from “幺,” and “二” from “两.” 3) English pronunciations within the audio should be rendered in corresponding Chinese characters or English. Specific cases include: all letters or words in URLs should be capitalized—for example, “www.abc.com” becomes “三W点A B C点com”; English words within the audio should be transcribed entirely in lowercase; English letters should be capitalized. Certain proper nouns or acronyms requiring all caps and spaces—such as “C E O,” “C C T V”—must follow these rules. 4) Annotation completeness must mirror actual pronunciation—no deletions allowed.

6. Data File Directory

Directory Structure

The directory tree reflects the hierarchical organization of the dataset.

│ .docx（Database Description）
│ README.txt（Overview）
│ ChangeLog（Data Version Information）
└─ data（Data Folder）
│├─wav.scp（Audio File List）
│├─trans.txt（Transcription List）
│├─spk_info.txt（Speaker Information List）
│├─wav（Audio Folder）
│├─D2166（Speaker Folder）
││ ││ ID2166W0001.wav（Audio File）