rkarhila/SIAK

"Say It Again, Kid!" (SIAK) Speech data collection##

Training data for pronunciation quality classifiers for childred learning English

Train set and test set in flac format.

File id key, fields separated by underscores (example: train001fifi05_609_t10892805_living-room.flac)

• Speaker key indicates train or test set, and a running number for speaker. speaker key is train001
• Native language: "fifi" for Finnish, "enuk" for UK English, "othr" for other. Native language fifi
• Age of speaker in years (if known). This speaker was 05 years old at the start of the recording period
• Sample number. This is the 609th sample spoken by the speaker. (Some kids really enjoyed contributing!)
• Seconds from first sample given. 10892805 seconds since first recording. This speaker contributed the samples over a 4 month period
• Targer utterance text with spaces etc replaced by dashes. Utterance to be spoken was "living room"

Release history

This data is derived from the data collected in the SIAK project 2014-2018, Participants agreed that their data can be published anonymously. Unfortunately the General Data Protection Regulation (GDPR) became effective before the data was ready for release, and the publication effort halted.

However the data was leased to an ill-fated startup that started operationsa few weeks before COVID-19 lockdowns. This collection is a derivation of the SIAK data with any strongly identifying metadata removed for use by the now bankrupt startup.

We were involved in collecting, storing and processing the data in the SIAK project and have gone through the speech samples in enough detail to be assured that the data can be regarded as non-personal and thus except from GDPR as it consists of only single words or very short utterance repetitions, making it next to impossible to identify a speaker.

Reima Karhila and Anna Smolander
SIAK project researchers and unlucky startup founders

---

license: cc-by-nd-4.0

We emphasize, that by no derivatives we mean that you cannot use the audio samples as part of any work that is not directly related to describing the dataset in a speech technology or scientific language learning context. You may include them in a scientific presentation when the context is clearly to present the original data and not to use the data in another fashion.

Commercial use of speech samples for building and evaluation of speech technology models is not prohibited.

If you publish work based on this dataset, please cite Karhila & al.: Pronunciation Scoring System Embedded into Children’s Foreign Language Learning Games with Experimental Verification of Learning Benefits, SLATE 2023.