#!/usr/bin/env bash

# fix segmentation fault reported in https://github.com/k2-fsa/icefall/issues/674
export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python

set -eou pipefail

stage=0
stop_stage=100
sampling_rate=24000
nj=32

dl_dir=$PWD/download

. shared/parse_options.sh || exit 1

# All files generated by this script are saved in "data".
# You can safely remove "data" and rerun this script to regenerate it.
mkdir -p data

log() {
  # This function is from espnet
  local fname=${BASH_SOURCE[1]##*/}
  echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
}

log "dl_dir: $dl_dir"

if [ $stage -le -1 ] && [ $stop_stage -ge -1 ]; then
  log "Stage -1: build monotonic_align lib"
  if [ ! -d vits/monotonic_align/build ]; then
    cd vits/monotonic_align
    python setup.py build_ext --inplace
    cd ../../
  else
    log "monotonic_align lib already built"
  fi
fi

if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
  log "Stage 0: Download data"

  # If you have pre-downloaded it to /path/to/LibriTTS,
  # you can create a symlink
  #
  #   ln -sfv /path/to/LibriTTS $dl_dir/LibriTTS
  #
  if [ ! -d $dl_dir/LibriTTS ]; then
    lhotse download libritts $dl_dir
  fi

  if [ ! -d $dl_dir/xvector_nnet_1a_libritts_clean_460 ]; then
    log "Downloading x-vector"

    git clone https://huggingface.co/datasets/zrjin/xvector_nnet_1a_libritts_clean_460 $dl_dir/xvector_nnet_1a_libritts_clean_460

    mkdir -p exp/xvector_nnet_1a/
    cp -r $dl_dir/xvector_nnet_1a_libritts_clean_460/* exp/xvector_nnet_1a/
  fi

fi

if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
  log "Stage 1: Prepare LibriTTS manifest"
  # We assume that you have downloaded the LibriTTS corpus
  # to $dl_dir/LibriTTS
  mkdir -p data/manifests
  if [ ! -e data/manifests/.libritts.done ]; then
    lhotse prepare libritts --num-jobs ${nj} $dl_dir/LibriTTS data/manifests
    touch data/manifests/.libritts.done
  fi
fi

if [ $stage -le 2 ] && [ $stop_stage -ge 2 ]; then
  log "Stage 2: Compute Spectrogram for LibriTTS"
  mkdir -p data/spectrogram
  if [ ! -e data/spectrogram/.libritts.done ]; then
    ./local/compute_spectrogram_libritts.py --sampling-rate $sampling_rate
    touch data/spectrogram/.libritts.done
  fi

  # Here we shuffle and combine the train-clean-100, train-clean-360 and
  # train-other-500 together to form the training set.
  if [ ! -f data/spectrogram/libritts_cuts_train-all-shuf.jsonl.gz ]; then
    cat <(gunzip -c data/spectrogram/libritts_cuts_train-clean-100.jsonl.gz) \
      <(gunzip -c data/spectrogram/libritts_cuts_train-clean-360.jsonl.gz) \
      <(gunzip -c data/spectrogram/libritts_cuts_train-other-500.jsonl.gz) | \
      shuf | gzip -c > data/spectrogram/libritts_cuts_train-all-shuf.jsonl.gz
  fi

  # Here we shuffle and combine the train-clean-100, train-clean-360
  # together to form the training set.
  if [ ! -f data/spectrogram/libritts_cuts_train-clean-460.jsonl.gz ]; then
    cat <(gunzip -c data/spectrogram/libritts_cuts_train-clean-100.jsonl.gz) \
      <(gunzip -c data/spectrogram/libritts_cuts_train-clean-360.jsonl.gz) | \
      shuf | gzip -c > data/spectrogram/libritts_cuts_train-clean-460.jsonl.gz
  fi

  if [ ! -e data/spectrogram/.libritts-validated.done ]; then
    log "Validating data/spectrogram for LibriTTS"
    ./local/validate_manifest.py \
      data/spectrogram/libritts_cuts_train-all-shuf.jsonl.gz
    touch data/spectrogram/.libritts-validated.done
  fi
fi

if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
  log "Stage 3: Prepare phoneme tokens for LibriTTS"
  # We assume you have installed piper_phonemize and espnet_tts_frontend.
  # If not, please install them with:
  #   - piper_phonemize:
  #       refer to https://github.com/rhasspy/piper-phonemize,
  #       could install the pre-built wheels from https://github.com/csukuangfj/piper-phonemize/releases/tag/2023.12.5
  #   - espnet_tts_frontend:
  #       `pip install espnet_tts_frontend`, refer to https://github.com/espnet/espnet_tts_frontend/
  if [ ! -e data/spectrogram/.libritts_with_token.done ]; then
    ./local/prepare_tokens_libritts.py
    touch data/spectrogram/.libritts_with_token.done
  fi
fi

if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
  log "Stage 4: Generate token file"
  # We assume you have installed piper_phonemize and espnet_tts_frontend.
  # If not, please install them with:
  #   - piper_phonemize:
  #       refer to https://github.com/rhasspy/piper-phonemize,
  #       could install the pre-built wheels from https://github.com/csukuangfj/piper-phonemize/releases/tag/2023.12.5
  #   - espnet_tts_frontend:
  #       `pip install espnet_tts_frontend`, refer to https://github.com/espnet/espnet_tts_frontend/
  if [ ! -e data/tokens.txt ]; then
    ./local/prepare_token_file.py --tokens data/tokens.txt
  fi
fi

audio_feats_dir=data/tokenized
dataset_parts="--dataset-parts all"  # debug "-p dev-clean -p test-clean"
if [ $stage -le 5 ] && [ $stop_stage -ge 5 ]; then
  log "Stage 5: Tokenize/Fbank LibriTTS for valle"
  mkdir -p ${audio_feats_dir}
  if [ ! -e ${audio_feats_dir}/.libritts.tokenize.done ]; then
    python3 ./local/compute_neural_codec_and_prepare_text_tokens.py --dataset-parts "${dataset_parts}" \
        --audio-extractor "Encodec" \
        --batch-duration 400 \
        --src-dir "data/manifests" \
        --output-dir "${audio_feats_dir}"
  fi
  touch ${audio_feats_dir}/.libritts.tokenize.done

  lhotse combine \
    ${audio_feats_dir}/libritts_cuts_train-clean-100.jsonl.gz \
    ${audio_feats_dir}/libritts_cuts_train-clean-360.jsonl.gz \
    ${audio_feats_dir}/libritts_cuts_train-other-500.jsonl.gz \
    ${audio_feats_dir}/cuts_train.jsonl.gz
  lhotse copy \
    ${audio_feats_dir}/libritts_cuts_dev-clean.jsonl.gz \
    ${audio_feats_dir}/cuts_dev.jsonl.gz
  lhotse copy \
    ${audio_feats_dir}/libritts_cuts_test-clean.jsonl.gz \
    ${audio_feats_dir}/cuts_test.jsonl.gz
fi