Dataset Card for MASRI_HEADSET_V2

Dataset Summary

TODO The MASRI-HEADSET CORPUS was created out of YouTube videos belonging to the channel of the University of Malta. It has a length of 1 hour and it is gender balanced, as it has the same number of male and female speakers.

Example Usage

The MASRI-HEADSET contains both a train and test split:

from datasets import load_dataset
masri_train = load_dataset("Bluefir/MASRI_HEADSET_v2", split="train")
masri_test = load_dataset("Bluefir/MASRI_HEADSET_v2", split="test")

### Supported Tasks
automatic-speech-recognition: The dataset can be used as a development/validation corpus when training a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER).
text-to-speech.
### Languages
The language of the corpus is Maltese.
## Dataset Structure
### Data Instances
```python
{
  'audio': {
    'path': '/HuggingFace/datasets/downloads/.../wav/MSRDV_F_08_DV_00005.wav', 
    'array': array([ 0.00140381,  0.00210571,  0.00189209, ..., -0.00210571,
       -0.00289917, -0.00372314], dtype=float32), 
    'sampling_rate': 16000
  }, 
  'speaker_id': 'F_08', 
  'gender': 'female', 
  'duration': 5.900000095367432, 
  'normalized_text': 'dawn ma kinux żminijiet sbieħ l-inkwiżizzjoni ddeċidiet li teqred lil kull min'
}

Data Fields

• audio (datasets.Audio) - a dictionary containing the path to the audio, the decoded audio array, and the sampling rate. In non-streaming mode (default), the path points to the locally extracted audio. In streaming mode, the path is the relative path of an audio inside its archive (as files are not downloaded and extracted locally).
• speaker_id (string) - id of speaker
• gender (string) - gender of speaker (male or female)
• duration (float32) - duration of the audio file in seconds.
• normalized_text (string) - normalized audio segment transcription

Data Splits

The corpus contains both a train and test split. TODO

Dataset Creation

Curation Rationale

TODO

The MASRI-DEV CORPUS (MDVC) has the following characteristics:

• The MDVC has an exact duration of 1 hours and 0 minutes. It has 669 audio files.
• The MDVC has recordings from 18 different speakers: 9 men and 9 women.
• Data in MDVC is classified by speaker. Therefore, all the recordings of each individual speaker are stored in one single directory.
• Data is also classified according to the gender (male/female) of the speakers.
• Every audio file in the MDVC has a duration between 3 and 10 seconds approximately.
• Audio files in the MDVC are distributed in a 16khz@16bit mono format.
• Transcriptions in MDVC are in lowercase. No punctuation marks are permitted except for dashes (-) and apostrophes (') due to their importance in Maltese orthography.

Source Data

Initial Data Collection and Normalization

Annotations

Annotation process

TODO Proper nouns and other words pronounced in languages other than Maltese (mainly from English, Italian, French and German) were transcribed in their respective orthographic system.

Who are the annotators?

TODO The audio transcription was performed by expert native speakers at the University of Malta in the Msida Campus.

Personal and Sensitive Information

TODO The dataset could contain names revealing the identity of some speakers; on the other side, the recordings come from a publicly repository (YouTube), so, there is not a real intent of the participants to be anonymized. Anyway, you agree to not attempt to determine the identity of speakers in this dataset. Notice: Should you consider that our data contains material that is owned by you and should therefore not be reproduced here?, please:

• Clearly identify yourself, with detailed contact data such as an address, telephone number or email address at which you can be contacted.
• Clearly identify the copyrighted work claimed to be infringed.
• Clearly identify the material that is claimed to be infringing and information reasonably sufficient to allow us to locate the material.
• Send the request to Carlos Mena Take down: We will comply to legitimate requests by removing the affected sources from the corpus.

Considerations for Using the Data

Social Impact of Dataset

TODO

Discussion of Biases

TODO

Dataset Curators

TODO

Licensing Information

TODO ? CC-BY-NC-SA-4.0. The copyright remains with the original owners of the video. As the data is taken from YouTube, we invoke the same argument of "fair use" as in the Voxlingua107 dataset, which is: "While YouTube users own the copyright to their own videos, using the audio in the videos for training speech recognition models has very limited and transformative purpose and qualifies thus as "fair use" of copyrighted materials. YouTube’s terms of service forbid downloading, storing and distribution of videos. However, the aim of this rule is clearly to forbid unfair monetization of the content by third-party sites and applications. Our dataset contains the videos in segmented audio-only form that makes the monetization of the actual distributed content extremely difficult."

Citation Information

TODO