Dne 19.května proběhlo slavnostní online vyhlášení výsledků soutěže Biosignal Challenge 2020 za účasti děkanů ČVUT FEL a 2.LF UK.


Nejlepšího celkového výsledku dosáhl a první cenu získal tým ve složení Ing. Ondřej Klempíř – Ph.D. student na katedře biomedicínské informatiky, FBMI ČVUT a Ing. David Příhoda – Ph.D. student ústavu informatiky a chemie, Vysoká škola chemicko-technologická v Praze. Cenu předal děkan 2. lékařské fakulty Univerzity Karlovy, prof. MUDr. Vladimír Komárek, CSc.


2. cenu získal tým ve složení Yeva Prysiazhniuk - studentka magisterského programuBioinformatikaoboruLékařská elektronika a bioinformatika” a Ilia Shipachev - student anglického magisterského programuPočítačové vidění a digitální obrazoboruOtevřená informatika“. Cenu předal děkan ČVUT FEL, prof. Mgr. Petr Páta, Ph.D.


3. cenu získali studenti magisterského programu Lékařská elektronika a bioinformatika ČVUT FEL Jan Ferkl a Samuel Maduda. Cenu předala Ing. Martina Mudrová, Ph.D. ze společnosti Humusoft, sponzora soutěže.


Čestné uznání Markétě Bařinkové - studentce Přírodovědecké fakulty UK a Filipu Šlapalovi - studentu magisterského programu Kybernetika a robotika předal prof. Ing. Roman Čmejla, CSc.



Odborná porota a organizační tým soutěže:

prof. Roman Čmejla, bc. Jan Vimr, dr. Jan Rusz,
dr. Radek Janča, dr. Michal Novotný, dr. Tereza Tykalová


Articulation rate estimation in children speech

1st price 800 USD

2nd price 400 USD

3rd price 200 USD


Articulation rate (AR) is a prosodic feature that indicates the number of spoken speech units per time. It is typically measured during connected speech where all types of pauses including silence, respiration and hesitations (such as /ah/, /um/, etc.) are excluded. Therefore, the articulation rate is mainly viewed as a representation of speech motor control since the linguistic effects are reduced.

Most researchers agree that AR can be affected by certain variables, which includes the length of utterance, locus of the word or phrase in the sentence and speaking context as well as speaking task such as reading, picture description, spontaneous speech, etc.

Different studies use different metrics to quantify AR. The most commonly used metrics are word per minute (WPM), syllable per second (SPS) and phoneme per second (PPS). The most suitable metric for AR estimation is SPS since syllables can be detected more easily than words or phonemes.

The goal of the Biosignal Challenge 2020 is to use the computing environment MATLAB to develop an algorithm for AR estimation in human speech signals by detecting the number of syllables (NOS) and measuring the duration of fluent speech (DFS), which requires excluding all types of pauses in each utterance.

Data and methods

From a database of 248 children’s utterances, 100 were randomly selected as a training dataset for your algorithms. The remaining utterance swill be used for validation. Utterances are from Czech children aged from 4 to 17 years at the time of the recording. The content of the recordings is a description of a picture portraying a young boy’s morning routine. The audio files are saved in .wav format with the sampling rate of 44.1 kHz.

AR was manually measured in all of these recordings by calculating the number of syllables and measuring the duration of fluent speech segments in each utterance. For this purpose, all parts with silence, respiration and filled pauses (hesitations such as /ah/, /um/, etc.) were excluded. The reference values reference_data.mat are saved in It contains vectors: true_syl and true_dur that represent manually measured values of number of syllables and duration of fluent speech respectively.

The task of the Biosignal Challenge 2020 is to develop an algorithm that calculates the number of syllables and the duration of fluent speech in each utterance. For your solution, you can use any possible method. The basic concepts related to the subject are matrix operations, energy, signal envelope, zero-crossing rate, autocorrelation, filtration, maximum detection, spectrum, power spectral density, spectrogram, spectral moments, basic statistics (mean, median, standard deviation, etc.).


The goal of this contest is to develop a MATLAB function in the form of function [syl, dur] = articulation_rate(sig, fs), where inputs are: sig – sampled sound, fs – sample rate, and outputs are: syl – number of syllables in the utterance, dur – duration of fluent speech in seconds. You can use the testing file test.p to evaluate your algorithm. It has to be in the same directory as your function articulation_rate.m, vector reference_data.mat as well as the training data. The outputs of the test file test.p are syllable detection accuracy and duration measurement accuracy. The numbers are calculated according to formulas:

where syl and dur are the calculated values, true_syl and true_dur are the reference values and and  are mean values of the reference vectors.

However, your algorithm will be validated on the different dataset (148 utterances from the same database as your training data). That is a common practice in algorithm evaluation to verify the robustness of the algorithm. Final score will be calculated as an average of syllable detection accuracy and duration measurement accuracy on the testing dataset.

Before you send us your algorithm, you should check whether it works with the mentioned test file to verify if your function is in the required form.


·         The Biosignal Challenge 2020 is an international competition for students from the European Union and the United States of America.

·         The teams can consist of 1 to 3 students

·         All algorithms must be programmed in the computing environment MATLAB.

·         The main function must be in the form:
function [syl, dur] = articulation_rate(sig, fs), where inputs are: sig – sampled sound, fs – sample rate, and outputs are: syl – number of syllables in the utterance, dur – duration of fluent speech in seconds.

·         Each team has to send their solution to email address up to May 11th, 2020

·         Your solution must contain:

o   the main function: articulation_rate.m

o   all extensions – functions, tools, data files, etc. (if necessary for the main function to operate)

o   the report in PDF which must contain: full names, university affiliation and degree of education of each member of the team, used MATLAB version, algorithm functionality, results on the training dataset, references


[1] Amir, O., Grinfeld, D.: Articulation Rate in Childhood and Adolescence: Hebrew Speakers, Language and speech, 54. 225-40, 2011


[2] Jong. N. H. de, Wempe T.: Praat script to detect syllable nuclei and measure speech rate automatically, Behavior research methods, vol. 41, no. 2, pp. 385–390, 2009

Files download