Datasets:
audio audioduration (s) 1.66 29.7 | text stringlengths 1 106 | lang stringclasses 2
values | accent stringclasses 10
values |
|---|---|---|---|
[unclear] | zh | cdo | |
Several versions exist, but they all involve "Earl Beardie" playing cards. | en | sg | |
Semiconductor chips do not usually exhibit "shmoo" shape plots. | en | us | |
There is also the City Island Seaside Trolley run by the Bronx Tourism Council. | en | sg | |
我觉得我也没什么会骄傲的啦,每天只这样过,做我自的事情,我这样,我也没特别去做什么会较特别,有tolong到人什做别样的事情。 | zh | cpx | |
Machan argued against animal rights in his widely reprinted paper Do Animals Have Rights? | en | hi | |
我今天和明天都要上課 | zh | tw | |
As James later explains to M, the lady made the advances. | en | us | |
Lazarus was born in Queanbeyan, New South Wales. | en | my | |
This branch is filled with friction composition in which a friction bar is embedded. | en | ur | |
Knob Noster State Park is nearby. | en | ur | |
Then the old quarries were used as landfills. | en | ur | |
Furthermore, the equations of motion impose that the Romans mass is constant. | en | ur | |
I wandered lonely as a cloud. | en | sg | |
準備吃午餐 | zh | tw | |
選定了心儀嘅機構先查詢個人實際年利率,咁會比較明智 | zh | hk | |
贷款第一笔资金去帮他们开始他们的生意这是一个很好的社区 | zh | cdo | |
In surprise finish, the final score was seven to five. | en | ur | |
Manager Dave Ross brought in Kris Taylor to form a management team. | en | us | |
Jef Raskin jokingly stated: Yes, we named our software after a bug. | en | ur | |
Was it a good idea? | en | sg | |
Double seaming uses rollers to shape the can, lid and the final double seam. | en | us | |
四 | zh | zh | |
“I am quite aware of that,” she replied. | en | ur | |
He did this by obscuring the reflector with a mask containing two vertical slits. | en | hi | |
則正確的說明了星球的大小 | zh | tw | |
Such conditions occur notably where the bedrock is limestone or recently formed volcanic lava. | en | my | |
They have two adult sons. | en | hi | |
Mechanical prediction approaches can incorporate clinical judgments, properly coded, in their predictions. | en | my | |
They said that they were forcefully vaccinated. | en | sg | |
二十一年去世。 | zh | zh | |
This leads to a break in the symmetry. | en | us | |
就變得超快的 | zh | tw | |
天馬行空 | zh | tw | |
In this case, another block is added to the message. | en | us | |
我们国家的学校是有教三个语言马来文华文和英文 | zh | cdo | |
Fuji Dream Airlines has its headquarters in Makinohara. | en | ur | |
In microprocessors, branch prediction permits avoidance of pipeline emptying at branch instructions. | en | hi | |
是內建的瀏覽器功能 | zh | tw | |
A group of people, one of them with a guitar, were sitting by the campfire and singing songs. | en | ur | |
And I am very very bitter about it... | en | sg | |
我好熱 | zh | hk | |
噢,我感到骄傲的因为我会吃,这久也许老了。哒,有孩儿,有老公,有儿,有孙,有女儿子,有外孙,[um]每次做年做节了叫他们回来齐[um]吃了讲讲了,多欢喜一晚上。 | zh | cpx | |
ok呃我们国家呃没很多语言很多那是三种而已ho英文国语华语 | zh | cdo | |
The Cadets attack the office and the rebellion is ended. | en | us | |
他们自称恰哈拉。 | zh | zh | |
我好熱 | zh | hk | |
The amenability property has a large number of equivalent formulations. | en | us | |
Don't overthink, just do it. | en | ur | |
說我自找 | zh | hk | |
After completing the fourth universe the game starts over. | en | my | |
我,它政府有帮助那非常不一样的那个话,因为读英文,读马来文,读华文,因为非常多人不相同,有辣子,有华人有马来油 | zh | cpx | |
杞人嘅朋友嘆咗一口氣 | zh | hk | |
醒少少啦 | zh | hk | |
八 | zh | hk | |
啊是啊小小的贷款啊可以帮助啊那小女孩做生意咯因为他们做的东西有时候只是做小小的做这手工什么也不需要做这么大的贷款吗so他们能够做这小小的东西开始啦啊不需要说一定要大大的东西开始大大的生意开始 | zh | cdo | |
阿列河畔贝赛。 | zh | zh | |
有個老人去左坪洲永安橫街食齋 | zh | hk | |
華頓商學院最受歡迎的談判課 | zh | tw | |
我能够大学毕业了这阵有讨到好的工咯。 | zh | cpx | |
马来西亚一个好就是它可以华语可以学英文可以学bahasa这是很好的一个国家它没有反对你们学多一点话但出国读书也比较方便 | zh | cdo | |
His poetry has features of a report, often with a distinct narrative story-line. | en | ur | |
Will you come with me? | en | my | |
Westman accepted and Rosenberger was voted out. | en | ur | |
正在做廉價勞工,收入少過做洗碗或保安員 | zh | hk | |
There is also an interchange with the Thousand Islands Parkway on the Ontario side. | en | us | |
我没有什么觉得骄傲的地方我我的人没什么很大的成功的地方都是很平常啦平常平常的过生活 | zh | cdo | |
It might be in this writer's Top Ten! | en | hi | |
二 | zh | zh | |
因為剛好不太冷又涼涼的 | zh | hk | |
我觉得我最骄傲的东西是[um]我从小个读书,自己会晓读书。免我的母亲去烦恼。 | zh | cpx | |
The station is owned by New River Community Church. | en | hi | |
有什么会让我骄傲我觉得不一定说很常对什么事情会觉得骄傲因为很多事情其实会做的出来做的会好就是不是我一个人做的也是有别人做的呃别人一起做的 | zh | cdo | |
你等我模擬 | zh | tw | |
Primary endosymbiosis involves the engulfment of a bacterium by another free living organism. | en | ur | |
勉強自己和別人好 | zh | tw | |
It is endemic to Western Tropical Africa. | en | hi | |
旗六公路 | zh | tw | |
This process is called differential erosion. | en | sg | |
富尔马诺夫出生在科斯特罗马省谢列达的一个农民家庭。 | zh | zh | |
Loved it. | en | ur | |
InfraGard focuses on the development, management and protection of critical infrastructure. | en | ur | |
no | en | ur | |
我觉得从小到大我希望我很欢喜的东西是我父亲和我的母亲给我们养到很好。好好做他人 | zh | cpx | |
What is it like? | en | my | |
The municipality consists of many islands. | en | hi | |
[Unclear]因为他们的事业有成,所以给我觉得很骄傲咯。还有,我们的家庭和谐。啊,较多是这样咯。 | zh | cpx | |
The British continued to use the site as a hospital. | en | ur | |
She also taught journalism at the University of California at Berkeley. | en | hi | |
呃,我觉得我今天若能够做一个骄傲的事情,能够帮助人较,人倒脚手一样东西,我就有心会特别快乐。[um]特别骄傲[um]这是我感到骄傲的地方。 | zh | cpx | |
渐渐行动不便 | zh | zh | |
三 | zh | hk | |
这个做生意啊不知道 | zh | cdo | |
老坑交流道 | zh | tw | |
More recent popular titles include "Prime Suspect," "The Forsyte Saga," "Sherlock," and "Downton Abbey". | en | hi | |
我觉得会骄傲的事情就是能够自自买车,买家,自生活自可以负责。这是那样到这久岁头很骄傲的事情咯,[um]甭讲要靠人呐,自能够靠自做工,赚钱自自生活,这是觉得讲我会很,很嘎骄傲的事情啦,到这久这样时,时间岁头啊。 | zh | cpx | |
Her intro later appeared on the David Holmes Essential Collection. | en | us | |
外出的人要小心! | zh | tw | |
黑身准裂腹鱼为辐鳍鱼纲鲤形目鲤科的其中一种。 | zh | zh | |
Norfolk Island is the only non-mainland Australian territory to have had self-governance. | en | hi |
CLD — English/Chinese Binary Speech Dataset (sample-size ablation)
Speech dataset for Convex Low-resource Accent-Robust Language Detection (CLD), covering the binary English vs. Chinese division. Provided at four training sample budgets as separate Hub configs for low-resource ablation studies.
Dataset description
We curate a dataset of multilingual voice transcriptions across high-resource languages and their low-resource sub-dialects. As a primary source of transcription data we use the Common Voice (v23) dataset (Ardila et al., 2020). We supplement this with several additional dialect datasets for regional speech variance:
- Singaporean English from the National Speech Corpus (NSC) — the first Singapore English corpus — provided through the Info-communications and Media Development Authority (IMDA) of Singapore. Singlish is selected because studies show it incurs particularly high error rates during voice transcription (Fong et al., 2002).
- The Lahaja dataset (Sanket et al., 2024), a benchmark comprising 12.5 hours of Hindi from 132 speakers across 83 Indian districts.
We normalize and augment all audio files via time stretching, volume gain, pitch shift, and augmented background noise with MUSAN (Snyder et al., 2015).
Schema
Each split is a datasets Dataset with
columns:
| column | type | description |
|---|---|---|
audio |
Audio(sampling_rate=16000) mono |
the speech clip, 16 kHz mono |
text |
string |
reference transcription |
lang |
string |
ISO-639-1 language code |
accent |
string |
accent / dialect label |
Binary setup
English and Mandarin are the two highest-resource languages in existing speech datasets, yet still display some of the lowest accuracy in language prediction for accented speech, due to the high variance of dialects and accents present in these two languages. For example, Whisper-Small achieves 100% accuracy on Midwestern English, drops to 91.8% on Wales-accented English, yet only 61.4% on Malaysian- accented English. We select 5 regional dialects per language and perform ablation studies on training sample sizes spanning 100 to 10,000 samples per language, splitting training samples equally across all accents.
- Languages (2):
en(English),zh(Chinese) - Configs (samples/language):
100,500,1000,10000 - Splits per config:
train/valid/test
How to use
from datasets import load_dataset
# pick a sample-size config: "100", "500", "1000", or "10000"
ds = load_dataset("williamhtan/cld-enzh-dataset", "1000")
print(ds)
sample = ds["test"][0]
print(sample["lang"], sample["text"])
Citation
If you use this dataset, please cite the CLD paper (ICML 2026) and the underlying corpora: Common Voice (Ardila et al., 2020), the National Speech Corpus (IMDA), Lahaja (Sanket et al., 2024), and MUSAN (Snyder et al., 2015).
- Downloads last month
- -