Datasets:
File size: 7,761 Bytes
1c4e6fd 654ac7b 1c4e6fd 654ac7b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 | ---
dataset_info:
features:
- name: user
dtype: string
- name: age
dtype: int64
- name: ethnicity
dtype: string
- name: gender
dtype: string
- name: phq1
dtype: float64
- name: phq2
dtype: float64
- name: phq3
dtype: float64
- name: phq4
dtype: int64
- name: phq5
dtype: int64
- name: phq6
dtype: int64
- name: phq7
dtype: int64
- name: phq8
dtype: float64
- name: phq9
dtype: int64
- name: phq
dtype: float64
- name: phq_2
dtype: float64
- name: gad1
dtype: float64
- name: gad2
dtype: float64
- name: gad3
dtype: float64
- name: gad4
dtype: float64
- name: gad5
dtype: float64
- name: gad6
dtype: float64
- name: gad7
dtype: float64
- name: gad
dtype: float64
- name: gad_2
dtype: float64
- name: duration
dtype: float64
- name: prompt_vad_cut_duration
dtype: float64
- name: prompt_vad_trimmed_duration
dtype: float64
- name: conversational_vad_cut_duration
dtype: float64
- name: conversational_vad_trimmed_duration
dtype: float64
- name: mos_pred
dtype: float64
- name: noi_pred
dtype: float64
- name: loud_pred
dtype: float64
- name: dis_pred
dtype: float64
- name: col_pred
dtype: float64
- name: bin_phq
dtype: int64
- name: bin_gad
dtype: int64
- name: split
dtype: string
- name: prompt
dtype: string
- name: state
dtype: string
- name: income
dtype: string
- name: english_preferred
dtype: string
- name: preferred_lang
dtype: string
- name: scores_anxiety
dtype: float64
- name: quantized_labels_anxiety
dtype: int64
- name: scores_depression
dtype: float64
- name: quantized_labels_depression
dtype: int64
- name: is_unbiased
dtype: bool
splits:
- name: validation
num_bytes: 4337865
num_examples: 8113
- name: test
num_bytes: 3743545
num_examples: 7034
download_size: 1717911
dataset_size: 8081410
configs:
- config_name: default
data_files:
- split: validation
path: data/validation-*
- split: test
path: data/test-*
license: apache-2.0
---
# Overview
This dataset includes audio-based model scores, clinical mental health labels, and demographic metadata for the validation and test sets used in the development of https://huggingface.co/KintsugiHealth/dam. Demographic statistics covering the training set as well are included in this model card.
The model was trained and evaluated on a large-scale speech dataset collected from approximately 35,000 individuals via phone, tablet, or web app, which corresponds to ~863 hours of speech data. The data was collected by asking each subject to speak for about 1 minute to various non-mental-health related prompts, e.g. talking about a favorite hobby or sport.
Ground-truth labels were derived from both clinician-administered and self-reported PHQ-9 and GAD-7 questionnaires, ensuring strong alignment with established clinical assessment standards.
The data consists predominantly of American English speech. However, a broad range of accents is represented, providing robustness across diverse speaking styles.
This data has been redacted for privacy purposes, so no personally identifiable information is included. In particular, the audio itself is excluded. In its place are model scores predicted by https://huggingface.co/KintsugiHealth/dam. Any audio streams containing less than the 30 seconds of speech needed to run this model have also been excluded from the dataset, but are included in the demographic statistics below.
While it is not possible to verify these scores by running the model directly, since the input audio is redacted, the authors hope the scores, accompanying labels, demographic data are useful for threshold tuning and independent research questions.
# Demographics
| | Train | Validation | Test | Total |
|:-------------------------------------------------------|:------------|:-------------|:------------|:------------|
| Number of unique subjects | 23,743 | 6,047 | 5,353 | 34,457 |
| Total number of audio files | 43,945 | 11,073 | 9,810 | 64,828 |
| Audio duration, s - Average (SD) | 56.4 (13.6) | 56.8 (13.8) | 56.1 (13.6) | 56.4 (13.7) |
| Audio duration, s - Median | 56.8 | 57.6 | 55.7 | 56.7 |
| Audio duration, s - Mode | 79.9 | 79.9 | 79.9 | 79.9 |
| Audio duration, s - Range | 30-229 | 31-231 | 31-176 | 30-231 |
| Audio duration, hr - Total | 688.5 | 174.7 | 152.9 | 863.2 |
| Age, y - Range | 18+ | 18+ | 18+ | 18+ |
| Age, y - Average (SD) | 45.2 (16.8) | 45.4 (16.8) | 45.5 (16.7) | 45.3 (16.8) |
| Age, y - Median | 42 | 43 | 43 | 42 |
| Age, y - Mode | 30 | 30 | 41 | 30 |
| Gender, % - Female | 57.5 | 57.8 | 57.8 | 57.6 |
| Gender, % - Male | 41.1 | 41 | 40.7 | 41 |
| Gender, % - Not specified | 0.4 | 0.4 | 0.3 | 0.3 |
| Gender, % - Other | 0.8 | 0.7 | 0.8 | 0.8 |
| Race/ethnicity, % - Asian or Pacific Islander | 5.9 | 6 | 6 | 6 |
| Race/ethnicity, % - Black or African American | 12.6 | 12.5 | 11.8 | 12.5 |
| Race/ethnicity, % - Hispanic or Latine | 17 | 17.2 | 17.7 | 17.2 |
| Race/ethnicity, % - Native American or American Indian | 0.9 | 0.8 | 0.8 | 0.8 |
| Race/ethnicity, % - Not specified | 2.9 | 2.9 | 2.9 | 2.9 |
| Race/ethnicity, % - Other or mixed race | 4.4 | 4.4 | 4.4 | 4.4 |
| Race/ethnicity, % - White | 56.2 | 56.1 | 56.5 | 56.2 |
| PHQ-9 score - Average (SD) | 8.7 (6.7) | 8.7 (6.8) | 8.8 (6.8) | 8.7 (6.7) |
| PHQ-9 score - Median | 7 | 7 | 8 | 7 |
| PHQ-9 score - Mode | 0 | 0 | 0 | 0 |
| PHQ-9 score - Range | 0-27 | 0-27 | 0-27 | 0-27 |
| GAD-7 score - Average (SD) | 7.3 (6.1) | 7.3 (6.2) | 7.4 (6.2) | 7.3 (6.2) |
| GAD-7 score - Median | 6 | 6 | 6 | 6 |
| GAD-7 score - Mode | 0 | 0 | 0 | 0 |
| GAD-7 score - Range | 0-21 | 0-21 | 0-21 | 0-21 |
# Threshold tuning
Illustrations of how to use this data to tune model thresholds are provided in the model card https://huggingface.co/KintsugiHealth/dam. |