Cnam-LMSSC/multilingual_librispeech_italian_phoneme

Multilingual LibriSpeech Italian Phoneme

Dataset Summary

This dataset is a curated version of the Italian subset of Multilingual LibriSpeech (MLS), enriched with a phonetic transcription column (phoneme).

The Laboratoire de Mécanique des Structures et des Systèmes Couplés (Cnam-LMSSC) created this version to facilitate research into Italian acoustic modeling, phoneme recognition, and speech synthesis. It builds upon the high-quality audio derived from LibriVox audiobooks found in the original MLS dataset.

Supported Tasks

• Phoneme Recognition: Training models to map audio features directly to IPA phonemes.
• Automatic Speech Recognition (ASR): Robust speech-to-text tasks using audiobook data.
• Text-to-Speech (TTS): Utilizing the high-quality, read-speech nature of LibriSpeech for synthesis.

Languages

The dataset contains audio in Italian (it).

Dataset Structure

Data Instances

A typical data point contains the audio, metadata about the speaker and chapter (derived from LibriVox), the original text, and the generated phonetic transcription.

{
  'id': '1234_5678_00001',
  'audio': {
    'path': '1234_5678_00001.flac',
    'array': array([-0.012, 0.045, ...], dtype=float32),
    'sampling_rate': 16000
  },
  'transcript': "presi li quattrocento fiorini d oro",
  'phoneme': "prezɪ lɪ kwatːrotʃɛnto fjorinɪ d̪ɔro",
  'speaker_id': 1234,
  'chapter_id': 5678,
}