eric/audita
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

Implemented deterministic transcript normalization before the LLM stages

Browse Source
This commit is contained in:
Eric Rakestraw
2026-04-22 07:22:40 -05:00
parent 445329de81
commit c1672af9b4
12 changed files with 719 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
README.md
View File

@@ -1,6 +1,6 @@
# Audita
Audita takes raw audio transcripts and uses an LLM to identify and fix misheard words, jargon, domain-specific terms, and conservative readability issues.
Audita takes raw audio transcripts, deterministically merges short same-speaker segments into speaking turns, and uses an LLM to identify and fix misheard words, jargon, domain-specific terms, and conservative readability issues.
## Development
@@ -47,6 +47,10 @@ Useful configuration can be supplied by CLI flag or environment variable:
- `AUDITA_MAX_RETRIES`, default `3`
- `AUDITA_GLOSSARY_MAX_LLM_PASSES`, default `3`, for total glossary correction passes
- `AUDITA_GRAMMAR_MAX_LLM_PASSES`, default `3`, for total grammar/readability correction passes
- `AUDITA_NORMALIZE_MAX_SEGMENT_GAP`, default `5.0`, for same-speaker gaps eligible for merging
- `AUDITA_NORMALIZE_ELLIPSIS_GAP`, default `2.0`, for same-speaker gaps that should be joined with ` ... `
- `AUDITA_NORMALIZE_MAX_SEGMENT_DURATION`, default `60.0`, for maximum merged segment duration
- `AUDITA_NORMALIZE_MAX_SEGMENT_TOKENS`, default `2048`, for maximum merged segment prompt payload size
- `AUDITA_WORK_DIR`, default `/tmp/audita`
`AUDITA_WORK_DIR` stores per-run diagnostics while processing. Successful runs clean up their run directory unless corrections are skipped; failed runs and skipped-correction runs preserve diagnostics for debugging.