Multilingual LibriSpeech Spanish Phoneme: 100K+ Audio-Phonetic Pairs

Between 100,000 and 1,000,000 Spanish audio segments and transcriptions derived from LibriVox audiobooks. Created by Cnam-LMSSC and updated in March 2026, it extends the Multilingual LibriSpeech (MLS) corpus with machine-generated phonetic transcriptions.

Use Cases

• Training acoustic models using the 'phoneme' column for precise speech-to-text alignment
• Developing Spanish Text-to-Speech (TTS) systems using the phonetic transcription sequences
• Fine-tuning Automatic Speech Recognition (ASR) models on Spanish-language audiobook data

Strengths

• Scale of 100,000 to 1,000,000 records
• Includes a dedicated 'phoneme' column for phonetic alignment
• Derived from high-quality LibriVox audiobook recordings

Limitations

• Phonetic annotations are machine-generated rather than human-verified
• Domain is restricted to audiobooks, which may not represent natural conversational speech

Provenance

Source

Laboratoire de Mécanique des Structures et des Systèmes Couplés (Cnam-LMSSC)

Collection Method

Curated from the Facebook Multilingual LibriSpeech subset with machine-generated phonetic annotations.

Freshness

Last updated March 2026.

Geography

Spanish-speaking