synth_bash_TTS_0.2.0

#!/bin/bash

# Neural text-to-speech (TTS) program using Mozilla TTS

# Clear output folder
echo "Clearing \"out\" folder..."
rm ./out/*

# Keep track of .wav output file number
count=0

# Loop over all input files
echo "Starting text-to-speech on files in \"in\" folder"
for file in ./in/*; do
    # Remove DOS-type newlines
    textUnix=$(tr --delete "\r" < "$file")
	# Remove hypenation of words due to the end of a line
	textToSynth1=$(echo "${textUnix}" | awk -v RS= '{gsub(/-\n/,""); print}')
    # Convert .txt input file to single-line text (added space to avoid gluing of titles with text)
    textToSynth2=$(echo "${textToSynth1}" | awk -v RS= '{gsub(/\n/," "); print}')
    # Convert single-line text to one sentence per line
    textToSynth3=$(echo "${textToSynth2}" | awk '{gsub(/\. /,".\n"); print}')
    # Remove anything between parentheses (i.e. citations and equation nubmers)
    textToSynth4=$(echo "${textToSynth3}" | awk '{gsub(/\([a-zA-Z0-9., ]*\)/,""); print}')
    # Remove anything between square brackets
    textToSynth5=$(echo "${textToSynth4}" | awk '{gsub(/\[[A-Z0-9.,]*\]/,""); print}')
	# Remove lines with less than 2 characters (incl. empty lines)
    textToSynth6=$(echo "${textToSynth5}" | awk 'length > 2')
    # Read each sentence and apply the text-to-speech model
    while read -r line; do
	# Split lines that are too long (the 1 assures that all lines are printed)
	# Words per line set to 12
	splitLine=$(echo "${line}" | awk '{for(i=10;i<NF;i+=10){$i=$i RS};gsub(RS FS,RS,$0)}1')
	while read -r sublines; do
	    tts --text "${sublines}" --model_name "tts_models/en/ljspeech/tacotron2-DDC" --out_path "out/"$count".wav"
	    count=$((count+1))
	done <<< "$splitLine"
    done <<< "$textToSynth6"
done

# This command makes sure we wait until the decoder is done
wait

# Run the script that merges the audio files
echo "Merging output files..."
python3 ./merging_wavs.py

echo "Done!"