How gdonaj/spoken-2-standard-evaluation works
This is a small command-line evaluator for spoken-to-standard language normalization. Users provide source text, standard references, and one or more system outputs; the script compares them and reports categorized errors and unchanged-word rates. Optional capabilities include approximate-randomization and bootstrap tests, plus LaTeX tables and HTML error visualizations. The implementation is a single script; the excerpts do not show all of its file-loading and dispatch logic, so the graph keeps internal responsibilities consolidated.
Main entry points: spk2std-evaluation.py Evaluation CLI
Components 10
Other components
- Evaluator user
- Source segments
- Standard references
- System outputs
- Evaluation CLI
spk2std-evaluation.py - Levenshtein
- Statistical tests
spk2std-evaluation.py - LaTeX report
- HTML error report
- Console results
Connections 9
- Evaluator user → Evaluation CLI: runs
- Source segments → Evaluation CLI: provides segments
- Standard references → Evaluation CLI: provides targets
- System outputs → Evaluation CLI: provides hypotheses
- Evaluation CLI → Levenshtein: aligns text
- Evaluation CLI → Statistical tests: tests systems
- Evaluation CLI → LaTeX report: writes optionally
- Evaluation CLI → HTML error report: writes optionally
- Evaluation CLI → Console results: prints results