Whisper transcriber (speech to text)

transcribe.sh

#!/usr/bin/env bash
# transcribe_chunks.sh — splits audio, uploads each chunk separately
# usage: transcribe_chunks.sh "input audio file"
# Requires OPENAI_API_KEY in environment.

set -euo pipefail

usage() {
  echo "Usage: $(basename "$0") \"input audio file\""
  echo "Requires OPENAI_API_KEY environment variable to be set."
  exit 1
}

[[ $# -eq 1 ]] || usage
[[ -n "${OPENAI_API_KEY:-}" ]] || {
  echo "Missing OPENAI_API_KEY env var" >&2
  exit 2
}

in="$1"
[[ -f "$in" ]] || {
  echo "No such file: $in" >&2
  exit 3
}

# ── prepare ────────────────────────────────────────────────────────────────
base="${in%.*}"
workdir="$(mktemp -d)"
trap 'rm -rf "$workdir"' EXIT

# ── split into ~9 minute chunks ─────────────────────────────────────────────
echo "Splitting into chunks..."
ffmpeg -loglevel error -i "$in" -f segment -segment_time 540 -c copy "$workdir/chunk-%03d.mp3"

chunks=("$workdir"/chunk-*.mp3)
[[ ${#chunks[@]} -gt 0 ]] || {
  echo "Splitting failed." >&2
  exit 4
}

# ── transcribe each chunk into its own file ─────────────────────────────────
echo "Transcribing ${#chunks[@]} chunks..."
n=1

for chunk in "${chunks[@]}"; do
  printf -v suf "%03d" "$n"
  out="${base}-${suf}.txt"

  size=$(stat --printf="%s" "$chunk")
  max_size=$((25 * 1024 * 1024))

  if ((size > max_size)); then
    echo "Error: chunk $chunk still too large (${size} bytes)." >&2
    exit 5
  fi

  curl -sS https://api.openai.com/v1/audio/transcriptions \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -F model=whisper-1 \
    -F response_format=text \
    -F file=@"$chunk" \
    --output "$out"

  echo "Chunk $n transcribed to $out"
  ((n++))
done