TechVoice: Human-Recorded Technical Speech for ASR Training, 205 Samples

205 human-recorded audio samples totaling 38 minutes and 43 seconds of speech focused on technical and developer vocabulary. The dataset, created by danielrosehill, is a work in progress with a target of 5 hours of audio and 50,000 words. It was last updated on November 26, 2025.

Use Cases

• Fine-tuning ASR models based on specialized technical vocabulary.
• Improving speech recognition accuracy for developer-focused content based on human recordings.
• Benchmarking ASR performance on niche, jargon-heavy speech data.

Strengths

• Contains 205 human-recorded audio samples, providing authentic speech data.
• Focuses on a specialized domain with technical and developer vocabulary.
• Current progress is tracked, showing 10,412 words and 38 minutes of audio collected.

Limitations

• Dataset is incomplete, currently at 12.9% of its 5-hour duration target.
• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

huggingface

Collection Method

Human-recorded speech samples.

Freshness

Last updated 2025-11-26 20:14:02; freshness should be verified.