Merge pull request #8 from BU-Spark/analysis-checkin1

[CORRECT ONE | IGNORE THE OTHER REQUEST] Analysis checkin

fall2024/PoC/README.md

@@ -0,0 +1,22 @@
+# Content
+This subdirectory consists mainly of analysis python notebooks. 
+
+# Metrics
+We are using:
+* WER - Calculated using **Levenschtein Distance Algorithm**
+* ROUGE Score - A combination of `Jieba` and `rouge_chinese` library
+
+# Attempts
+We have attempted to draw relationship between **stuttering count** and performance by the followings:
+* plotting **stuttering count** to performance metrics => no pattern found
+* plotting **audio length** to performance metrics => no pattern found
+
+# Notes
+* We are thinking about the indirect performance analysis assuming shorter **audio length** implies less **stuttering count** => no evidence for assumption
+
+# Issues
+* Some `pandas` parsed dataframe has mismatched ground truth to the original ground truth transcription data file provided. We are checking if the parsing has gone wrong, otherwise removing those transcripts for analysis.
+
+# Upcoming Agenda
+* Normalized analysis => merging **stuttering count** and **audio length** into new parameter **frequency**
+* Process the falsely parsed transcription