Contact for any inquiries Email: [email protected] | Web: http://deeplyinc.com/ | Tel: (+82) 70-7459-0704
Pairs of Korean speakers reading a script with 3 distinct text sentiments (negative, neutral, positive), with 3 distinct voice sentiments (negative, neutral, positive), are recorded. The recordings took place in 3 different types of places, which are an anechoic chamber, studio apartment, and dance studio, of which the level of reverberation differs. And in order to examine the effect of the distance of mic from the source and device, every experiment is recorded at 3 distinct distances with 2 types of smartphone, iPhone X, and Galaxy S7.
There were two individuals in a group, and each group was distinguished by a unique key(subject ID). Two speakers(speaker a, speaker b) were facing each other 1.4m apart from each other (0.7m from the middle line). They read the allocated scripts alternating between speaker a and b, as if they were talking to each other.
| Recording environment | Studio Apartment(moderate reverb), Dance studio(high reverb), Anechoic Chamber(no reverb) |
|---|---|
| Device | iPhone X(iOS), Samsung Galaxy S7 |
| Recording distance from the source | 0.4m, 2.0m, 4.0m |
| Volume(Sample) | ~ 290(~ 3) hours, ~ 190,000(~ 2,000) utterances, ~ 107(~ 0.5) GB |
| Format | wav/h5(16/44.1kHz, 16-bit, mono) |
| Language | Korean |
| Studio Apartment | Dance studio | Anechoic Chamber |
|---|---|---|
 |
 |
 |
Refer to the dataset descriptions in 'docs' for detailed description and statistics of the full set of the dataset.
The dataset is a subset(approximately 1%) of a much bigger dataset which were recorded under the same circumstances as these open source samples. Please contact us([email protected]) for the pricing and licensing.
The illustrations below are the statistics about the Deeply Korean Read Speech Corpus. The first three are from the sample dataset, And the others are from the full dataset. To attain more insight about the dataset, please refer to the detailed description in 'docs' and 'Korea_Read_Speech_Corpus.json' in 'Dataset'.
The sample is a partial set of recordings from a single subject group(sub1001), which consists of 20-year-old female(speaker a) and 49-year-old female(speaker b).
├── Dataset
│ ├── AirbnbStudio
│ │ ├── sub100100a00000.wav
│ │ └── ...
│ ├── AnechoicChamber
│ │ ├── sub100120a00000.wav
│ │ └── ...
│ ├── DanceStudio
│ │ ├── sub100110a00000.wav
│ │ └── ...
│ └── Korean_Read_Speech_Corpus.json
└── docs
├── Deeply\ Korean\ Read\ Speech\ Corpus_Eng.pdf
└── Deeply\ Korean\ Read\ Speech\ Corpus_Kor.pdf
Korean_Read_Speech_Corpus.json
{'AirbnbStudio':
{'sub100100a00000': {'text_sentiment': 0,
'voice_sentiment': -1,
'subjectID': 'sub1001',
'speaker': 'a',
'age': 20,
'sex': 0,
'noise': 0,
'location': 0,
'distance': 0,
'device': 0,
'text': '저 식당 음식이 정말 맛있나 봐요.',
'text_code': 'aa0',
'rms': 0.024304501712322235,
'length': 2.71825},
...
},
...
}
Text sentiment: {-1: negative, 0: neutral, 1: positive}
Vocie sentiment: {-1: negative, 0: neutral, 1: positive}
Subject ID: Unique 'sub + 4-digit' key allocated to each subject group
Speaker: unique key allocated to each indiivdual in the subject group.
Sex: {0: Female, 1: Male}
Noise: {0: Noiseless, 1: Indoor noise, 2: Outdoor noise, 3: Both indoor/outdoor noise}
Location: {0: Studio apartment, 1: Dance studio, 2: Anechoic chamber}
Distance: {0: 0.4m, 1: 2.0m, 2: 4.0m}
Device: {0: iPhone X, 1: Galaxy S7}
Text: the content of the script
Text code: unique key allocated to each line of the script
Rms: Root mean square value of the signal
Length: length of the signal in secods
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
- Deeply Nonverbal Vocalization Dataset
- A human nonverbal vocal sound dataset by Deeply Inc.
- Deeply Parent-Child Vocal Interaction Dataset
- The interaction of pairs of parent and child(reading fairy tales, singing children’s songs, conversing, and others).Recorded in 3 types of places, at 3 distinct distances, with 2 types of smartphone.
Tel: (+82) 70-7459-0704
Web: http://deeplyinc.com/
Email: [email protected]