Mandarin Speech Corpus with 218 Speakers and 85 Hours

85 hours of emotion-neutral Mandarin speech recordings from 218 native speakers, comprising 88,035 utterances. The corpus is designed for training multi-speaker Text-to-Speech systems and includes auxiliary speaker attributes such as gender, age group, and native accent labels.

Use Cases

• Train a multi-speaker TTS model using 88,035 utterances and speaker-specific attributes like gender and age group.
• Develop accent-aware speech synthesis systems leveraging the native accent labels provided for each speaker.
• Build speaker verification or identification models using the high-fidelity recordings from 218 distinct individuals.
• Analyze phonetic or prosodic patterns across different demographic groups defined by the gender and age group metadata.

Strengths

• Contains 85 hours of high-fidelity Mandarin speech audio.
• Includes metadata for 218 native speakers with attributes like gender and age group.
• Comprises a large collection of 88,035 individual utterances.

Limitations

• Recordings are described as emotion-neutral, limiting use for expressive or emotional speech synthesis.
• The corpus focuses solely on Mandarin, which may not be suitable for multilingual TTS research.
• No information is provided on the recording environment or potential background noise levels.

Provenance

Source

Beijing Shell Shell Technology Co., Ltd. (AISHELL).

Collection Method

Recordings of 218 native Chinese Mandarin speakers.

Freshness

Last updated on the platform in January 2024.

Geography

Presumably China, given the focus on native Mandarin speakers.