Emotional Speech Corpus
Enter your email below for pricing and additional information. For a faster reply, please contact - contact@deeplyinc.com
Summary
Pairs of speakers reading a script with 3 distinct text emotion (sentiments; negative, neutral, positive), with 3 distinct voice emotion (sentiments; negative, neutral, positive), are recorded. The recordings took place in 3 different types of places, which include an anechoic chamber, studio apartment, and dance studio, of which the level of reverberation differs. In order to examine the effect of the distance of mic from the source and device, every experiment is recorded at 3 distinct distances using 2 types of smartphone, iPhone X, and Galaxy S7.
All participants are Korean. 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 center line). They read the provided 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 (contact@deeplyinc.com) for the pricing and licensing.
Featured Emotional Speech Data Sample
-
Sentence with negative meaning
-
Sentence with neutral meaning
-
Sentence with positive meaning
Sentence meaning
"Emotional description is important, but this is just violence"
"The food in that restaurant looks really good"
"I wish there were more dramas like this. There was a lot to think about, the visuals were beautiful, and the characters were all attractive."
-
Negative voice reading
-
Neutral voice reading
-
Positive voice reading
"The food in that restaurant looks really good"
"The food in that restaurant looks really good"
"The food in that restaurant looks really good"
Data sample according to the recording device
-
iPhone X (iOS)
-
Samsung Galaxy S7 (Android)
Data sample according to recording distance
-
Between sound source and devices: 0.2m
-
Between sound source and devices: 2.0m
-
Between sound source and devices: 4.0m
Data sample according to recording space
-
Recorded in room
-
Recorded in dance studio
-
Recorded in anechoic chamber
Dataset statistics
The illustrations below are the statistics about the Deeply Korean Read Speech Corpus. The first three graphs are from the sample dataset and the others are from the full dataset. To gain more insight regarding 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).
Structure
├── 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.',
'rms': 0.024304501712322235,
'length': 2.71825},
...
},
...
}
How to decode
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 indvidual 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 seconds
License
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Enter your email below for pricing and additional information. For a faster reply, please contact - contact@deeplyinc.com
Office : E02, Space Sallim 2F, 10, Noryangjin-ro, Dongjak-gu, Seoul, Republic of Korea
Tel : +82 70-7459-0704
E-mail : contact@deeplyinc.com
Copyright © Deeply, Inc. All rights reserved.