AudioChinese dialects

Hakka Dialect Database

About this dataset

1. Product Overview

The Hakka dialect speech database contains a total of 513 hours of recordings. The recorded language is the Hakka dialect, and the recordings were made in Meizhou, China. A total of 411 speakers were invited to record using their mobile phones in quiet environments. This database has undergone professional proofreading, achieving a word accuracy rate of over 96%, making it suitable for research in areas such as voiceprint recognition and speech recognition.

2. Recording Corpus

Transcription Results

The read-aloud portion of this database consists of conversational texts, including everyday interactions such as greetings among friends and family, communication between colleagues, and customer service inquiries. Each speaker recorded a different number of texts, with the following distribution:

Number of Sentences · Number of Speakers · Percentage (100%)

0–250 · 166 · 40.39%

250–500 · 69 · 16.79%

500–1K · 87 · 21.17%

1K–1.5K · 46 · 11.19%

1.5K–2K · 23 · 5.60%

2K–2.5K · 6 · 1.46%

2.5K–6K · 14 · 3.41%

Corpus Processing

 De-identification: Removal of politically sensitive, personally private, or pornographic and violent content.  Removal of special symbols such as <, >, [, ], ~, /, \, =, except for numeric strings.  Exclusion of content containing languages other than Chinese and English.  Standardization of formatting.

3. Speaker Information

Gender Distribution

The database includes a total of 411 speakers: 67 males and 344 females.

Gender · Male · Female · Total

Percentage · 16.3% · 83.7% · 100%

Age Distribution

The age distribution of the speakers is as follows: A (16–25 years old); B (26–40 years old); C (41–50 years old); D (51+ years old).

· Age Group · Number of Speakers · Percentage

A · 16–25 years old · 84 · 20.43%

B · 26–40 years old · 251 · 61.07%

C · 41–50 years old · 60 · 14.6%

D · 51+ years old · 16 · 3.9%

Total · 411 · 100% ·

4. Data Recording

Recording Environment

A quiet indoor setting, free from echoes and other voices.Recording Equipment

Common smartphones such as Apple, Huawei, Xiaomi, OPPO, etc., were used. The data was stored in 16 kHz, 16-bit WAV format.Recording Method

For the read-aloud portion, speakers recorded indoors using a smartphone app in close‑talk mode, reading naturally without rote memorization.Recording Duration Distribution

Speakers randomly selected texts and read them naturally. Individual recording sessions lasted no more than 20 hours but at least one minute. The specific duration distribution is as follows:

Duration (seconds) · Number of Speakers · Percentage (100%)

0–1K · 100 · 24.33%

1K–2K · 80 · 19.46%

2K–4K · 78 · 18.98%

4K–6K · 53 · 12.90%

6K–10K · 51 · 12.41%

10K–40K · 49 · 11.92%

5. Speech Data Transcription

Transcribers transcribed audio into text, striving to ensure consistency between written content and spoken pronunciation. General guidelines are as follows: 1) The transcription must exactly match the spoken audio—no extra words, missing words, or errors. 2) Numbers should be written in Chinese characters, e.g., “one two three,” rather than “123.” Pay attention to distinguishing “one” from “yao,” and “two” from “liang.” 3) English pronunciations in the audio should be rendered in corresponding Chinese characters or left in English. Specific cases include: All letters or words in URLs should be capitalized—for example, if the audio says “www,” transcribe it as “three W dot A B C dot com.” English words within the audio should be transcribed entirely in lowercase. English letters appearing in the audio should be transcribed entirely in uppercase. For certain proper nouns written in all caps, maintain capitalization without spaces—for example, CEO, NBA, etc. 4) The completeness of annotations must align with actual pronunciation; no omissions allowed. 5) Except for traditional expressions (such as “a week,” “a year,” “the second day of the lunar month”), proprietary terms (like “ten years sharpening a sword”), and vague phrases (such as “one or two hundred,” “three or four”), all other data should be expressed in Arabic numerals.

6. Data File Directory

Directory Structure

The directory tree structure is as follows:

└─DOC (Text Description Files)
││├─content.txt (Transcription List)
││├─spk-info.txt (Speaker Information Table)
└─SPEECHDATA (Audio Folder)
││├─K001 (Speaker ID)
│││├─K001W000001.wav (Audio File)