Runs whisper cpp along with ffmpeg to record voice to transcribe. Tested on mac m2

whispercpp_loop

#!/usr/bin/env bash

# This script runs the speech-to-text loop from the commandline. It records to given file till you press CTRL-C or 30 seconds has passs`ed
# and then it transcribes the recording and prints the result. It then starts recording again and so on.

# Install whisper
:'
git clone git@github.com:ggerganov/whisper.cpp.git
cd whisper.cpp
# Download whisper open source model base version
# Good enough for englisg
bash ./models/download-ggml-model.sh base.en # Download 
make
'
# Usage: ./run_stt_loop.sh
# Dependencies: ffmpeg
# Install: sudo apt-get install ffmpeg

set -e

mkdir -p /tmp/stt
REC_LOCATION=/tmp/stt/rec
REC_ITERATION=0
REC_FILE=

log () {
  ts=$(date +"%Y-%m-%d %H:%M:%S")
  echo "$ts: $1"
}

# interrupt on space
trap 'transcribe' SIGINT
trap 'echo "Exiting after cleanup..."; exit 0' SIGTERM

transcribe() {
  log "Transcribing recording.wav"
  ./main $REC_FILE -nt -otxt 
  log "Transcription complete:"
  cat $REC_FILE.txt 
  rm $REC_FILE
  log "Waiting 5 seconds before recording again"
  sleep 5
  run_loop
}

# List recording devices and let user choose device unless recoarding device id is given as first argument
re='^[0-9]$'
if ! [[ $1 =~ $re ]] ; then
  device=$(ffmpeg -f avfoundation -list_devices true -i "" 2>&1 | grep -E "\[AVFoundation.*\] \["|sed -E 's,[AV.*] ,,'|fzf --reverse|grep -o '\[.\]'|sed 's,\[,,'|sed 's,\],,')
else
  device=$1
fi  
log "Device selected: $device"

# Start recording and transcribing loop
run_loop() {
  # Record to file
  REC_ITERATION=$((REC_ITERATION+1))
  REC_FILE=$REC_LOCATION$REC_ITERATION.wav
  log "Recording to file recording.wav. Press CTRL+C to transcribe or else this will transcribe in 30 seconds"
  rm -f  $REC_FILE
  ffmpeg -f avfoundation -i ":$device" -t 30 -ar 16000 $REC_FILE -hide_banner -loglevel error
  transcribe
}

run_loop

rm -f  $REC_FILE