UCLASS

A partly-annotated corpus of audio examples of stuttered speech. Corpus can be used in training models for recognizing stuttered speech and the non-fluent speech that all speaker produce occasionally.

User can access recordings of speakers who stutter and background details about these speakers and the conditions in which the recordings were made. The recordings are available in various formats. The main two sets of recordings were made in normal speaking conditions and the final one was made when the sound of the speaker's voice was altered as he or she spoke.

The audio data have been prepared so that they can be downloaded and played on a PC fitted with a sound card (WAV and MP3 formats) or so they can be handled and analyzed by freeware available on the web (note that MP3 format is compressed to 32 kilobits per second). The three freeware packages are CHILDES, PRAAT and SFS.

Transcription data (available for some files in UCLASS Release One and Two)

     For CHILDES. Transcriptions were prepared using an in-house script which gives a CHAT file (MacWhinney, 1995) and the corresponding audio files are also available in WAV format. The WAV files have been linked to the CHAT files. The analysis software, CLAN, has options that allow PRAAT programs to process the associated WAV files. Other information that CHAT files include in their headers is available in the ACCESS files described above for Releases One and Two.

CHAT files + WAV = CHILDES

     For PRAAT. The transcriptions have been converted to PRAAT TextGrids. PRAAT provides acoustic analysis facilities for dealing with WAV files that are also available in the directory.

TextGrids + WAV = PRAAT

     For SFS. SFS stands for Speech Filing System and, as its name suggests, it files sources of information together (e.g. audio and transcriptional). There is a wide range of utilities available in SFS. For instance, the audio waveforms and aligned transcriptions can be displayed and manipulated using some of the SFS utilities. There is an SFS file corresponding to each of the recordings. Most of these just contain the audio file. Some of the remaining ones have phonetic transcriptions that have been aligned manually against the audio file. Separate orthographic and phonetic transcriptions are available for some of the files as text files. The phonetic transcriptions are in JSRU format (see “How we transcribe” under shared resources on the main page of this site). It is straightforward to translate these to other phonetic formats.

The data and software are freely available to anyone for research and teaching purposes subject to the conditions that they a) acknowledge the source of the data, b) indicate that data collection was supported by the Wellcome Trust.