Apply for the Challenge
Register an account
Fill out the application
Submit the application
|Ranking||Name||Team Name||Grade||Time of the best result submitted|
For this challenge, a database covering China's 10 major dialects will be open. Challengers shall take both or either of two tasks with different difficulties.
In each task, challengers are required to build a system that automatically identifies and assorts the audio files with different durations (≤3s for the first task and >3s for the second task) provided in the challenge. The final ranking will be decided based on the classification accuracy of your system.
Training set and development set is open for challengers' use but the test set is not open to the public.
1)Apply for the challenge and download the training set and development set on the official website from Mar. 22
2)Submit your work for an online test from May 15 to Jun. 19 (the test ranking will not be used for the Preliminary round).
1)The best test result that you get during the period of Jun. 20 to Jul. 19 (Not including the test ranking) will be your preliminary result.
1)Those who are selected into this part shall download the newly-added training set and development set, debug the algorithm and submit their works (only once a day) through the official website.
2)The best result that you get during the period of Jul. 20 to Sept. 19 will be your Semifinal result.
3)The list of Top 32 will be published at 11:00 am on Sept. 19 on the official website. And all of them will get a free ticket to iFLYTEK 1024 Developer Festival and be awarded a certificate of participation.Top 8 among the Top 32 will be qualified to the Final.
Support from iFLYTEK (29/09-19/10)
1)Top 8 will get support from senior scientists at iFLYTEK Research.
2)Top 8 shall submit a PPT document for the Final.
1)Top 8 will be qualified to the Final at iFLYTEK 1024 Developer Festival.
2)Top 8 shall prepare for the final with 10-minute speech (with a PPT) and 5-minute Q&A.
3)The first place, second place, third place and winning prize winner (1 place) will be selected on Oct. 24, 2018. Your performance in both the Final and Semifinal will be considered for our judgment.
Six dialects in Preliminary: Changsha Dialect, Hebei Dialect, Nanchang Dialect, Shanghai Dialect, Fujian Dialect and Hakka. For each dialect, there will be 6-hour audio data covering 40 speakers.
Training set: there will be 6000 audio files in each dialect (200 for every speaker) and 30 speakers (15 males and 15 females).
Development set & Test set: for each dialect, there will be 5 speakers (2 females and 3 males for development set while 3 females and 2 males for test set). Each set is divided into two categories according to the duration of the audio file (≤3s for the first task and >3s for the second task). Every speaker in each task has 50 audio files.
The phonetic sequence annotation of the corresponding text to each speech is also provided.
Note: You are only required to finish the second task (>3s) in this challenge.
About the data details, please see the Figure 1.
|Dateset in Preliminary round||Training set||Development set||Test set|
|Dialect code||Categories||Accent area||Speakers||Sentences for each speaker||Amount of sentences||Speakers||Sentences for each speaker||Amount of sentences||Speakers||Sentences for each speaker||Amount of sentences||Speakers||Sentences for each speaker||Amount of sentences|
|changsha||Changsha||Changsha and its surrounding areas||30||200||6000||5||50||250||5||50||250||5||100||500|
|nanchang||Nanchang||Nanchang and its surrounding areas||30||200||6000||5||50||250||5||50||250||5||100||500|
|shanghai||Shanghai||Shanghai and its surrounding areas||30||200||6000||5||50||250||5||50||250||5||100||500|
|kejia||kakka||Mei County/Meizhou/Huiyang/ surrounding areas||30||200||6000||5||50||250||5||50||250||5||100||500|
|minnan||Fujian||Xiamen/Zhangzhou/Quanzhou/ surrounding areas||30||200||6000||5||50||250||5||50||250||5||100||500|
Figure 1 Data Details（Note: the highlighted font is the new data set for the Semifinal）
There are no restrictions on how the challenge system is built. All machine learning methods can be used.
The system can be a combination of various methods, such as voting method.
Two completely independent systems can be used for two different tasks or for both semi-final and preliminary rounds.
Since your work submitted is tested through an offline manner, the response time of your competition system is not required.
Challengers shall submit their systems by themselves for the test set of this competition is not open.
Please be noted that:
a)Your name, the lead author, the task name and the classification accuracy in training set and development set shall be marked when submitting your system(s).
b)A paper or an instruction to specifically explain your system(s) shall be provided when submitting your work in Semifinal. The constitution of your system(s), the training methods and corresponding parameters shall be included in your explanation. Certainly, it will be better if you can provide the source code.
c)The ranking list with the classification accuracy of your work will be published and updated at 11:00 am every day on the official website.
All the tests will be performed on 64 bit Linux CPU server. Works created under any other OS systems instead of 64 bit Linux or any GPU or FPGA based platform will not be accepted.
You are NOT permitted to:
a)Use any other data other than those provided by iFLYTEK in this Challenge.
b)Change or modify the phonetic sequence annotation.
c)Process the dateset by any other means, such as speech endpoint detection to the dataset.
You are permitted to:
a)Conduct machine simulation or noise addition only using our training dataset.
b)Use all the information contained in the datasets.