The dataset contains synthesized replicas of sustained vowels /A/ and /I/ performed by healthy controls, patients with Parkinson’s disease, multiple system atrophy and progressive supranuclear palsy. The dataset can be used as a reference for evaluation of pitch detectors, detectors of modal fundamental frequency, and detectors of subharmonics.Coding system

Each recording is named by a unique alphanumeric code in the format Uvxy, where U means abbreviation of the group (HC = healthy control, PD = Parkinson’s disease, MSA = multiple system atrophy, PSP = progressive supranuclear palsy) in upper case characters, v is the numeric identifier of the subject within the group, x denotes type of vowel (a = vowel /A/, i = vowel /I/), and y is the number of repetition. The part of the code of U and v uniquely determine each speaker, whereas x and y determine speaker’s recordings.

Recordings

All recordings are briefly described in table dataset.csv. All files of each record (see records.zip) are identified by the corresponding code and suffix. Suffix describe type of the file and is separated from the code by underscore. Naming of the files is illustrated on the record HC8a1. The code describes first repetition of the vowel /A/ performed by healthy speaker HC8. The record HC8a1 consists of following files:HC8a1.wav = waveform of the synthesized replica. This is the reference signal used for the evaluation. Parameters of jitter, shimmer and harmonic to noise ratio (HNR) can be found in dataset.csv.HC8a1_clean.wav = waveform of the synthesized replica without added noise. We provide this signal to make the model more versatile. Authors may add a different kind of noise to this signal or manipulate with HNR. Note that that both signals required normalization prior to writing into wav-file. Original scaling factor between HC8a1 and HCa1_clean can be determined from total power of signals and reference HNR value.HC8a1_LF.wav = sample of the glottal pulse used for the synthesis.HC8a2_impulses.csv = list of impulses’ locations in seconds and corresponding amplitudes. The position of pulses was corrected to match with the beginning of glottal pulse, i.e., first sample of the signal HC8a1_LF.wav begins at each of these positions. The jitter and shimmer listed in dataset.csv were are median values. Jitter and shimmer by other definitions can be calculated from positions and amplitudes of pulses provided by this file. HC8a1_subharmonics.csv = list of subharmonic intervals described by the start time in seconds and end time in seconds. Corresponding index of amplitude modulation expressed as SHR in percent can be found in the table dataset.csv. When no subharmonic was determined by the supervised parameterization, no file was included for the speaker and SHR in the dataset.csv was set to zero [1].

[1] Note that the supervised detection had lower sensitivity due to senzitivity of pitch trace in PRAAT, so the occurrence of subharmonics in synthesized data is much lower than in the original dataset analyzed by automated segmentation. This is not a problem because subharmonics were synthesized only at the given interval - this illustrates why it is important to detect subharmonics in other way than by pitch.

I would be grateful if you cite my two following papers:

a self-collected Parkinson's speech dataset

The NeuroVoz dataset emerges as a pioneering resource in the field of computational linguistics and biomedical research, specifically designed to enhance the diagnosis and understanding of Parkinson's Disease (PD) through speech analysis. This dataset is distinguished as the first of its kind to be made publicly available in Castilian Spanish, addressing a critical gap in the availability of linguistic and dialectical diversity within PD research.

Compiled from a cohort of 112 participants, including 54 individuals diagnosed with PD and 58 healthy controls, the NeuroVoz dataset offers a rich compilation of speech recordings. All PD participants were recorded under medication (ON state), ensuring consistency and reliability in the speech samples collected. The dataset is meticulously curated to include a variety of speech tasks—ranging from sustained vowel phonations and diadochokinetic (DDK) tests to 16 structured listen-and-repeat utterances and spontaneous monologues. The inclusion of both manually transcribed listen-and-repeat tasks and Whisper-automated transcriptions for monologues underscores our commitment to data accuracy and usability.

Encompassing 2,977 audio files, the NeuroVoz dataset provides an extensive repository, averaging 26.88 +- 3.35 recordings per participant, making it an invaluable asset for researchers seeking to explore the nuances of PD-affected speech. The dataset's structure and composition facilitate a multifaceted analysis of speech impairments associated with PD, offering insights into phonatory, articulatory, and prosodic changes.

In contributing to the body of knowledge with the NeuroVoz dataset, we invite the scientific community to engage with this dataset, explore the specific speech characteristics of PD in Castilian Spanish speakers, and advance the field of PD diagnosis through innovative speech analysis techniques.

Paper explaining the database: https://arxiv.org/abs/2403.02371
Zenodo repository with the database: https://zenodo.org/records/13647600
Github repo on how to use the dataset: https://github.com/BYO-UPM/Neurovoz_Dababase