Improvement in Automatic Speech Recognition of South Asian Accent Using Transfer Learning of DeepSpeech2

Algorithm 1

Dataset preprocessing algorithm.

	Read CSV of common voice dataset
	Choose “Pakistani,” “Indian,” “Dutch,” and “Sri-Lankan” accent rows from country column
	Select “filename” columns from CSV file
	Create a dictionary for JSON file writing
	for i in filename do
	flacConvert(i)
	SilenceRemover(i)
	Duration(i)
	Finding transcript with respect to audio file
	for lineIndex in range(x) do
	transcript = split[lineIndex]
	Write JSON file
	data = {“audio_filepath”: “1.FLAC,” “duration”: duration, “text”: transcript}
	with open(‘json-other-train.txt,” “a”) as outfile:
	json.dump(data, outfile)
	end for
	end for
	flacConvert(i) (Convert mp3 -> FLAC)
	Calculate audio file length
	Calculate audio file sample rates
	Calculate duration = audio file length/sample rate
	Calculate loudness
	Calculate peak amplitude
	Splitting audio file into chunks with silence > 0.5 second (considering it silent if quieter than −16 dBFS)
	for chunk in enumerate (chunks) do
	List.Append(chunk)
	end for