Japanese Speech Recordings From Mobile Phones In Multiple Environments

Featuring 210,000 commonly used Japanese sentences recorded by 799 local speakers. Audio was captured in quiet indoor places, streets, and restaurants using mainstream Android phones and iPhones, with a text transcription error rate below 5%.

Use Cases

• Train an automatic speech recognition (ASR) model on 210,000 Japanese sentences recorded in multiple acoustic environments.
• Analyze speech characteristics and pronunciation variations across the 799 local speaker cohort.
• Fine-tune a text-to-speech model using audio recordings paired with their corresponding written sentence transcriptions.
• Study the impact of recording environment (quiet indoor, street, restaurant) on speech clarity and model performance.

Strengths

• Contains 210,000 Japanese sentence recordings, providing substantial audio data.
• Includes recordings from 799 distinct local Japanese speakers.
• Audio was captured in diverse real-world environments: quiet indoor places, streets, and restaurants.
• Text transcription error rate is reported to be less than 5%.

Limitations

• The dataset is described as a sample of a larger paid dataset, so the full scope is not available.
• No information is provided on the demographic distribution (age, gender) of the 799 speakers.
• The specific recording devices are listed only as 'mainstream Android phones and iPhones', lacking detailed microphone specifications.

Provenance

Source

Nexdata

Collection Method

Recorded by 799 Japanese locals reading sentences on mobile phones in various locations.

Time Range

null

Freshness

null

Geography

Japan (speakers are Japanese locals)