improve wording

lijp22 · Aug 7, 2021 · bcd5bfd · bcd5bfd
1 parent 06d7944
commit bcd5bfd
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 4 deletions.
diff --git a/HOW_TO_SUBMIT.md b/HOW_TO_SUBMIT.md
@@ -17,7 +17,7 @@ Conceptually, for leaderboard to re-produce and benchmark submitters' ASR system
 * runtime resources (e.g. model, config, cloud-api credentials)
 * a program that can decode local audio list
 
-So the main purpose of leaderboard pipeline is to formalize above aspects down to a concrete contract. Let's start with `submission model dir`:
+So the main purpose of leaderboard pipeline is to formalize above aspects down to a concrete specification. Let's start with `submission model dir`:
 ```
 jiayu@ubuntu: tree ./sample_submission_model_dir
 

diff --git a/README.md b/README.md
@@ -23,9 +23,8 @@ As above figure shows, SpeechIO leaderboard serves as an ASR benchmarking platfo
   - open for people who are willing to share/publish their ASR models
 
 3. an open benchmarking pipeline:
-  - defines a simplest-possible contract on the format of test sets, recognition result etc.
-  - defines a simplest-possible benchmarking interface for model submitters
-  - a fully automated pipeline to perform `prepare` -> `recognize` -> `process` -> `evaluate`, leaderboard users don't need to write code.
+  - defines a simplest-possible specification on recognition interface, the format of input test sets, the format of output recognition results.
+  - As long as model submitters conform to this specification, a fully automated pipeline will take care of the rest (e.g. data preparation -> recognition invocation -> text post processing -> WER/CER/SER evaluation)
 
 With SpeechIO leaderboard, anyone can benchmark/reproduce/compare performances with arbitrary combinations between test-set zoo and model zoo, by simply filling a request form [example form](requests/sample_request.yaml)