Multilingual speech dataset with urgency tone annotation
Description
The following dataset is a collection of multilingual audio recordings in a controlled environment, divided according to the tone of urgency used in a hypothetical emergency scenario.
The dataset contains 200 audio files in 5 different languages: English, French, German, Italian, and Dutch. There are 40 audio files for each language, divided equally (20 and 20) between those labelled as urgent and those labelled as non-urgent. Urgent audio files are recordings of speech in simulated emergency situations in which the speaker adopts a tone defined as urgent. Non-urgent audio files are recordings of general content in which the speaker uses a neutral tone of voice.
The English audio files are produced using artificial intelligence-based text-to-speech software.
The remaining audio files for each language are recorded by the same person (i.e. 4 people), who are native speakers of the respective language.
test_for_thresholds are the italian audio used to set the prosodic thresholds between urgent and not_urgent.
noisy_audio are the original 200 recordings augmented by applying five degradation conditions: additive white Gaussian noise at 5, 15, and 30 dB Signal-to-Noise Ratio (SNR), environmental background noise, and room impulse response reverberation
All audio files are provided in uncompressed WAV format with the following characteristics:
- Codec : pcm_s16le
- Sample rate: 48 kHZ
- Channels: 2 (stereo)
Structure of the dataset:
dataset:
data:
languages:
urgent:
audio_xx_urgent.wav
transcription_language.txt
not_urgent:
audio_xx_not_urgent.wav
transcription_language.txt
test_for_thresholds:
xu.wav
noisy_audio:
languages:
urgent:
urgent_AWGN_5
urgent_AWGN_15
urgent_AWGN_30
urgent_ENV
urgent_RIR
Transcription_xx_xx.txt
not_urgent:
urgent_AWGN_5
urgent_AWGN_15
urgent_AWGN_30
urgent_ENV
urgent_RIR
Transcription_xx_xx.txt