# DSP Final Project

## Final report
https://docs.google.com/document/d/1VgmEAUuor8mXLuRDJTeQiFOgc1tSf0N_bjBmeiaH4-Q/edit

## TODO
- Convert each sec ... done
- Convert wave file to spectrogram ... done
- Find a female voice data set ... done
- 剪掉voice前面的slicence ... done
- 先這樣

## Reference
https://github.com/andabi/deep-voice-conversion
https://github.com/robmsmt/KerasDeepSpeech
https://medium.com/@jonathan_hui/speech-recognition-deep-speech-ctc-listen-attend-and-spell-d05e940e9ed1

* THCHS-30 資料集論文
https://arxiv.org/pdf/1512.01882.pdf

* CTC Loss 論文
https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.75.6306&rep=rep1&type=pdf

* CTC Loss 圖片
https://distill.pub/2017/ctc/

[DeepSpeech2 on PaddlePaddle](https://github.com/PaddlePaddle/DeepSpeech)

## text to speech
[Tacotron-2](https://github.com/Rayhane-mamah/Tacotron-2)

## voice segmentation
[pyAudioAnalysis](https://github.com/tyiannak/pyAudioAnalysis)
https://github.com/tyiannak/pyAudioAnalysis/wiki/5.-Segmentation

[inaSpeechSegmenter](https://github.com/ina-foss/inaSpeechSegmenter)

## Data set
`organize_data_info.py`: generate data set info , store to .csv file
`run_wav2mel.py`: <filename list> <output dir>

Voice data(.wav): https://drive.google.com/drive/folders/1m6euURTOeQhEcJI4wguDgvvWy8xptvp9?usp=sharing

Generate file list:

$ for f in ./*.wav; do echo "$f" >> mylist.txt; done