README.md

metadata

task_categories:
  - automatic-speech-recognition
language:
  - en
tags:
  - microbiology
  - science
  - synthetic
  - kokoro
  - speech

Kokoro Science and Microbiology STT Synthetic Dataset

Synthetic English scientific speech generated with hexgrad/Kokoro-82M.

Current contents

• 8,995 utterances
• 20.000 measured hours
• 16 kHz mono embedded FLAC audio
• text: desired ASR transcript
• tts_text: pronunciation-friendly text supplied to Kokoro

Corpus generations

• v1_10h_original: original approximately 10-hour release
• v2_20h_expansion: second tranche bringing the dataset to 20 measured hours

v2 expansion focus

• mM vs mL and µM vs µL unit/context contrasts
• dilution notation and CFU enumeration
• microbiology organisms and selective media
• antimicrobial susceptibility and MIC language
• molecular biology and qPCR terminology
• laboratory QC and reporting language

Categories

• microbiology: 3,795
• units_concentrations: 1,032
• quantitative_science: 695
• enumeration: 606
• laboratory_methods: 590
• molecular_biology: 527
• antimicrobial_susceptibility: 455
• chemistry: 382
• quality_science: 254
• biochemistry_general_science: 197
• unit_contrasts: 167
• general_science: 136
• scientific_language: 84
• laboratory_qc: 75

Synthetic speech is useful for domain adaptation, but real-human scientific speech should remain an independent validation target.