Welcome to DLHLP2020 Homework 1, end-to-end speech recognition for Bopomofo characters. 

## Homework description
[Slide link](https://docs.google.com/presentation/d/1C2a1_V91EHjy3sgxy3rKWqU6hCIOOsbMWuOhYCZW5jQ/edit?usp=sharing)

## Kaggle Upload Deadline
2020/03/22 (Sun.) 23:59 (UTC+8)

## Evaluation Metric

**Mean [Levenshtein Distance](https://en.wikipedia.org/wiki/Levenshtein_distance)** calculated for each sentence in char-level.

Levenshtein distance between two character sequence (predicted sequence v.s. target sequence) is the minimum number of single-character edits (insertions, deletions or substitutions) required to change one word into the other.
\[ LevDistance = N_{ins}+N_{del}+N_{sub} \]
p.s. **Word-Error Rate (WER)** is normalized Levenshtein Distance (to target sequence length), a more common evaluation metric for ASR.
\[ WER = \frac{N_{ins}+N_{del}+N_{sub}}{N_{target-length}} \]
Submission Format
All student should submit your answer in CSV format with
,
as delimiter
. Submission files should contain two columns:
id
and
answer
.
id
should be a six-digits number (e.g.
009001
according to
009001.wav
audio file. The answer to the wav file should be put in
answer
column. There should
NOT
be any space between the
answer
and delimiter. Sorting according to `id` is not necessary.
There is an example format:
id,answer
009001,ㄩˇ ㄧㄣ ㄅㄧㄢˋ ㄕˋ
009002,ㄩˇ ㄧㄣ ㄅㄧㄢˋ ㄕˋ
...,...
009999,ㄩˇ ㄧㄣ ㄅㄧㄢˋ ㄕˋ
010000,ㄩˇ ㄧㄣ ㄅㄧㄢˋ ㄕˋ