xuhuan0217 / audio-data-pre-processing

Audio-data-pre-processing

---

Audio pre-processing, especially, signal preprocessing related topics is not what I familiar with. My work and experience are mostly about CV and NLP and other data analysis areas. But I am happy to learn new things, I used to learn FFT when I am a undergraduate student but never really use it. Redo the FFT part givea me some inspiration about how NN learns new features, you can find it on fft notebook. Mel-scale is completely new to me.

And I am not sure the connection between part 1 and 2, as I dont know how the 41 wav files and meta data can bu used further.

---

0. data collection

• requests+beautifulsoup analyze soundwave website and extract necessory metadata including track_name,author,genre,play_count,download_count,comments_count,like_count,audio_url
• soundwave api for automatic audio downloading.
• ./data folder is too big, I will upload to Google drive and share later

1. audimetadata

• show meta data as request.

2. fft

• DFT: discrete Fourier Transform
• FFT: recurisive 1D FFT
• FFT_vectorized: non-recurisive vetorized FFT
• NN DFT: NN based DFT
• Different domain visualization contains animation, if you cannot open on github please download or check the colab version.

3. Spectrogram and mel-scale

• STFT
• Spectrogram with FFT and STFT
• Mel-scale filter bank
• Mel-scale Spectrogram

---