Haitian Creole Speech Recordings For ASR And TTS

A collection of over 7 hours of voice recordings from nearly 100 Haitian Creole speakers, intended for building Automatic Speech Recognition (ASR) and Text-to-Speech (TTS) systems. It is primarily sourced from the CMU Haitian Creole Speech Recognition Database and has been filtered and organized for training Mozilla's DeepSpeech model.

Use Cases

• Train an ASR model on over 7 hours of Haitian Creole voice recordings to transcribe spoken language.
• Develop a TTS system using voice recordings from nearly 100 speakers to synthesize Haitian Creole speech.
• Filter and organize audio data specifically for compatibility with the Mozilla DeepSpeech architecture.

Strengths

• Over 7 hours of recorded audio data provides a substantial base for speech model training.
• Includes voice samples from nearly 100 different speakers, offering speaker diversity.
• Data is pre-filtered and organized for a specific use case with Mozilla DeepSpeech.

Limitations

• The dataset size of over 7 hours is relatively small for training robust, general-purpose speech models.
• Specific audio quality metrics, speaker demographics, and recording conditions are not detailed.
• The data is sourced from a single primary database, which may limit linguistic or acoustic variety.

Provenance

Source

Primarily the CMU Haitian Creole Speech Recognition Database.

Collection Method

Voice recordings filtered and organized for model training.

Time Range

null

Freshness

null

Geography

null