The Cleft Dataset

A dataset of ultrasound and audio recorded with children with cleft lip and palate

The cleft dataset is a collection of ultrasound tongue imaging and audio data, gathered from children with cleft lip and palate by a research speech and language therapist working in a hospital environment.


We recorded data with 39 English-speaking children, but only 29 gave consent to share their data. These are 11 female speakers and 18 male, aged 7-11 years.

01M M 10 5 10.42 BCLP no
03F F 5 1 5.83 UCLP no
05M M 10 3 10.75 UCLP no
06M M 9 8 9.75 UCLP no
07M M 9 2 9.75 UCLP no
09M M 9 8 9.33 UCLP yes
11M M 4 5 4.42 CP no
12F F 5 1 5.42 CP yes
14F F 5 0 5.33 BCLP no
15F F 4 9 4.75 UCLP yes
16M M 9 7 9.33 UCLP no
17F F 4 4 4.42 BCLP no
18F F 5 1 5.25 UCLP no
19F F 3 9 3.58 CP yes
20F F 7 5 7.75 CP no
21M M 9 1 9.5 CP no
24M M 6 5 6.33 CP yes
25M M 4 1 4.33 CP no
26M M 4 4 4.67 BCLP no
28F F 8 9 8.58 BCLP no
30F F 7 7 7.42 CP no
31F F 5 4 5.42 CP no
32M M 5 8 5.5 UCLP no
33M M 6 4 6.42 UCLP yes
34M M 5 3 5.25 CP yes
35M M 3 7 3.42 UCLP no
36M M 5 0 5.58 BCLP no
37M M 7 0 7.58 BCLP no
39M M 7 0 7 CP yes

Cleft Types

Data type Description
CP cleft palate only
UCLP unilateral cleft lip and palate affecting one side of the lip and palate
BLP bilateral cleft lip and palate affecting both sides


Each child recorded an "Assessment" session, and two children recorded a "Therapy" session.

Data Types

Core data types:

Data type Description
wav speech waveform
ult raw ultrasound data
param ultrasound parameters
txt prompt text with date/time of utterance recording

Hardware synchronisation failed for this dataset. We release the parameter File as exported from the AAA software. In [1] we provide automatically predicted synchronisation offsets.

Additional data:

Data type Description
slt_labels manual annotation from SLT, when available. See [2] for details
probe_direction_labels a label for each utterance indicating whether the probe was in a coronal position (cor) or midsagittal right or let (sag_right, sag_left)

SLT Labels are available in Praat's TextGrid format, and probe_direction_labels is a csv file.

File IDs

Individual recordings are indexed for each session according to their recording times. See the prompt text file for recording date/time.

Each file ID also includes a prompt type identifier. See Data for details.


[1] Eshky, A.,Cleland, J., Ribeiro, M. S., Renals, S. Automatic audiovisual synchronisation for ultrasound tongue imaging. (Under revision).

[2] Cleland, J., Lloyd, S., Campbell, L., Crampin, L., Palo, J.-P., Sugden,E., Wrench, A., & Zharkova, N. (2020). The impact of real-time ar-ticulatory information on phonetic transcription: ultrasound-aidedtranscription in cleft lip and palate speech. Folia Phoniatrica etLogopaedica, 72, 120–130.