Transcribe audio files to plain text using OpenAI Whisper API. Compresses via ffmpeg (mono, 22kHz, 64kbps) and trims if over 25MB.

transcribe

#!/usr/bin/env bash
set -euo pipefail

if [ -z "${OPENAI_API_KEY:-}" ]; then
  echo "Error: OPENAI_API_KEY is not set" >&2
  exit 1
fi

if [ $# -ne 1 ] || [ ! -f "$1" ]; then
  echo "Usage: transcribe <audio-file>" >&2
  exit 1
fi

input="$1"
tmpfile=$(mktemp /tmp/transcribe-XXXXXX.mp3)
trap 'rm -f "$tmpfile"' EXIT

max_bytes=$((25 * 1024 * 1024))

echo "Compressing audio (mono, 22kHz, 64kbps)..." >&2
ffmpeg -y -i "$input" -ac 1 -ar 22050 -b:a 64k "$tmpfile" -loglevel error >&2

filesize=$(stat -f%z "$tmpfile")
echo "Compressed to $(($filesize / 1024 / 1024))MB." >&2

if [ "$filesize" -ge "$max_bytes" ]; then
  duration=$(ffprobe -v error -show_entries format=duration -of csv=p=0 "$tmpfile")
  bytes_per_sec=$(echo "$filesize $duration" | awk '{printf "%.0f", $1 / $2}')
  max_dur=$(echo "$bytes_per_sec" | awk '{printf "%.0f", (24 * 1024 * 1024) / $1}')
  echo "Still too large, trimming to ${max_dur}s..." >&2
  ffmpeg -y -i "$input" -ac 1 -ar 22050 -b:a 64k -t "$max_dur" "$tmpfile" -loglevel error >&2
  filesize=$(stat -f%z "$tmpfile")
  echo "Trimmed to $(($filesize / 1024 / 1024))MB." >&2
fi

echo "Sending to Whisper API..." >&2
curl -s https://api.openai.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -F file="@$tmpfile" \
  -F model="whisper-1" \
  -F response_format="text"