Treble10-Speech: 16 kHz ASR Data with 10 Simulated Room Environments

Treble10-Speech is an automatic speech recognition (ASR) dataset featuring 16 kHz audio files generated by convolving LibriSpeech data with high-fidelity room-acoustic simulations. Created by Treble Technologies and updated in November 2025, the collection includes between 1,000 and 10,000 records across 10 distinct furnished room environments. The dataset provides speech samples with reverberation times ranging from 0.17 to 0.84 seconds.

Use Cases

• Training ASR models to handle reverberation in specific domestic environments like bathrooms or bedrooms
• Evaluating speech enhancement algorithms against varied reverberation times (0.17-0.84 s)
• Benchmarking acoustic simulation accuracy using the Treble10-RIR based convolutions

Strengths

• High-fidelity room-acoustic simulations from Treble10-RIR
• 10 distinct furnished room environments with specific volumes
• Controlled reverberation times (RT60) between 0.17 and 0.84 s
• Standardized 16 kHz sampling rate for ASR compatibility

Limitations

• Small record count (1,000 to 10,000 records)
• Synthetic convolution rather than real-world physical recordings
• Limited to 10 specific indoor room configurations

Provenance

Source

Treble Technologies, based on OpenSLR LibriSpeech and Treble10-RIR simulations

Collection Method

Synthetic convolution of clean speech with simulated room impulse responses

Freshness

Last updated November 2025.