Multilingual Speech Corpus with 12,800 Audio Samples

Featuring 12,800 balanced audio samples in WAV format and related transcriptions from 18 speakers. It is assembled from multiple sources including VCTK, LJSpeech, m-ailabs, and SIWIS, covering languages such as English, French, German, Luxembourgish, and Portuguese.

Use Cases

• Train a text-to-speech model using the 12,800 WAV audio files and their corresponding transcriptions.
• Perform speaker identification or verification across the 18 distinct speakers included in the corpus.
• Develop multilingual speech synthesis systems leveraging the dataset's coverage of English, French, German, Luxembourgish, and Portuguese languages.
• Analyze acoustic features or prosody from the 16 kHz sampled audio files for speech processing research.

Strengths

• Contains 12,800 balanced audio samples, providing a substantial base for model training.
• Includes data from 18 different speakers, offering speaker diversity.
• Multilingual coverage spans at least five languages: English, French, German, Luxembourgish, and Portuguese.
• Audio files are provided in a consistent WAV format sampled at 16,000 Hz.

Limitations

• The specific distribution of samples per language or speaker is not detailed, which may obscure potential class imbalances.
• The dataset is a composite from multiple sources with potentially varying recording conditions and quality.
• Last update was in July 2022, which may not reflect the latest advancements or data availability in the field.

Provenance

Source

Assembled from VCTK, LJSpeech, m-ailabs, and SIWIS datasets.

Collection Method

Custom assembly from multiple existing open speech corpora.

Freshness

Last updated on 2022-07-12.