SolomidHero

wav_path = str(REPO_PATH / "tests/accent_test_sample/split/accent17/accent17_a17_14s.wav")
wav_16, sr16 = librosa.load(wav_path, sr=16000)
wav_22, sr22 = librosa.load(wav_path, sr=22000)
print(wav_22.shape, wav_16.shape)

# test1: sr=16khz, win=640, hop=160
from new_data.preprocessing.batch_processors.spectrogram_creator import (
    SpectrogramCreator
)
from new_data.train_data_containers import Batch

SolomidHero

#!/bin/sh

# archive: tar + pigz
tar -cf split.tar.gz -I pigz dir/
# unarchive: tar + pigz
tar -xf split.tar.gz -I pigz -C ./


# archive: zip
zip -rq -0 split2.zip split/

SolomidHero

import subprocess
import logging
import datetime
from pathlib import PosixPath
import time
from torch.utils.tensorboard import SummaryWriter


check_frequency = datetime.timedelta(minutes=5)
aws_url = 's3://<bucket>/<path>'

SolomidHero

#! /bin/bash

# common utils
apt update && apt-get install -y sudo git vim wget unzip rubberband-cli sox
sudo apt-get install -y libsndfile1-dev libopenblas-base
pip install gdown librosa

# conda environment for Montreal-Forced-Aligner
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -b -p ~/miniconda

SolomidHero

# speaker embeddings by AdaIN-VC speaker encoder
# ref: https://github.com/cyhuang-tw/AdaIN-VC
#
# 1. clone repo above
#   $ git clone https://github.com/cyhuang-tw/AdaIN-VC
#   $ cd AdaIN-VC
# 2. download pretrained checkpoint (by repo authors) and unzip it
#   $  gdown --id 1d_llv1qaCpPjioReh4AT8K_-qqG2zoIx
#   $  unzip -qo /content/vctk_model.zip
# 3. change pathes in this script and run it (with `python3 to_adain_emb.py`)

SolomidHero

# Least Squares GAN loss

def adversarial_loss(scores, as_real=True):
  if as_real:
    return torch.mean((1 - scores) ** 2)
  return torch.mean(scores ** 2)


def discriminator_loss(fake_scores, real_scores):
  loss = adversarial_loss(fake_scores, as_real=False) + adversarial_loss(real_scores, as_real=True)

SolomidHero

#!/bin/bash
# Word count of file using map reduce
# Args:
#   - file
#   - n_workers

map() {
  local file=$1
  local from=$2
  local to=$3

SolomidHero

// modular constant
const int m = 1e10 + 7;
// array: x <-> inverse to x
int invs[maxN];

// number of combinations C_n^k
// uses `invs` array as memory
int c_n_k(const int n, const int k) {
  if (k > n) return 0;

SolomidHero

// Author: Dmitry Sadykov
// For simplicity, next one-liner is useful to copy paste into several files:
// $ curl https://gist.githubusercontent.com/SolomidHero/dc1bb22679e5e0310399b443738ecfb8/raw/b16d6a2987efa77c6320679da3fb0f65cbefd594/template.cpp | pbcopy; for i in {a..f}; do pbpaste > $i.cpp; done

#include <iostream>

using namespace std;

using ll = int64_t;