egs/ami/ASR/prepare.sh

#!/usr/bin/env bash

set -eou pipefail

stage=-1
stop_stage=100
use_gss=false   # ❌ Disable GSS (not needed for IHM-only)

dl_dir=$PWD/download

. shared/parse_options.sh || exit 1

mkdir -p data
vocab_size=500

log() {
  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"

# -----------------------------
# Stage 0: Download AMI (IHM only)
# -----------------------------
if [ $stage -le 0 ] && [ $stop_stage -ge 0 ]; then
  log "Stage 0: Download AMI IHM data (~70h)"

  if [ ! -d $dl_dir/amicorpus ]; then
    lhotse download ami --mic ihm $dl_dir/amicorpus
  fi
fi

# -----------------------------
# Stage 1: Prepare AMI IHM manifests
# -----------------------------
if [ $stage -le 1 ] && [ $stop_stage -ge 1 ]; then
  log "Stage 1: Prepare AMI IHM manifests"

  mkdir -p data/manifests

  lhotse prepare ami \
    --mic ihm \
    --partition full-corpus-asr \
    --normalize-text kaldi \
    --max-words-per-segment 30 \
    $dl_dir/amicorpus data/manifests/
fi

# -----------------------------
# Stage 4: Compute fbank features (AMI IHM only)
# -----------------------------
if [ $stage -le 4 ] && [ $stop_stage -ge 4 ]; then
  log "Stage 4: Compute fbank features for AMI IHM"

  mkdir -p data/fbank
  python local/compute_fbank_ami.py

  log "Creating final IHM training cuts"
  gunzip -c data/manifests/cuts_train_ihm.jsonl.gz | \
    gzip -c > data/manifests/cuts_train_all.jsonl.gz
fi

# -----------------------------
# Stage 6: Dump transcripts for BPE training (IHM only)
# -----------------------------
if [ $stage -le 6 ] && [ $stop_stage -ge 6 ]; then
  log "Stage 6: Dump IHM transcripts for BPE"

  mkdir -p data/lm

  gunzip -c data/manifests/ami-ihm_supervisions_train.jsonl.gz | \
    jq '.text' | sed 's:"::g' > data/lm/transcript_words.txt
fi

# -----------------------------
# Stage 7: Prepare BPE-based language resources
# -----------------------------
if [ $stage -le 7 ] && [ $stop_stage -ge 7 ]; then
  log "Stage 7: Prepare BPE-based lang"

  lang_dir=data/lang_bpe_${vocab_size}
  mkdir -p $lang_dir

  echo "<eps> 0" > $lang_dir/words.txt
  echo "!SIL 1" >> $lang_dir/words.txt
  echo "<UNK> 2" >> $lang_dir/words.txt

  cat data/lm/transcript_words.txt | \
    grep -o -E '\w+' | sort -u | \
    awk '{print $0, NR+2}' >> $lang_dir/words.txt

  num_words=$(wc -l < $lang_dir/words.txt)
  echo "<s> ${num_words}" >> $lang_dir/words.txt
  num_words=$(wc -l < $lang_dir/words.txt)
  echo "</s> ${num_words}" >> $lang_dir/words.txt
  num_words=$(wc -l < $lang_dir/words.txt)
  echo "#0 ${num_words}" >> $lang_dir/words.txt

  ./local/train_bpe_model.py \
    --lang-dir $lang_dir \
    --vocab-size $vocab_size \
    --transcript data/lm/transcript_words.txt

  if [ ! -f $lang_dir/L_disambig.pt ]; then
    ./local/prepare_lang_bpe.py --lang-dir $lang_dir
  fi
fi