An ML application that transcribes audio files into text and may be done in English, French, Spanish, and Arabic.
Transcribing audio files into text generally is a time-consuming task, especially if you’ve gotten a big volume of audio data that should be transcribed. Fortunately, machine learning (ML) algorithms may be used to automate this process and make it much faster and more efficient.
In this text, I’ll discuss an ML application that I built which may transcribe audio files into text in 4 different languages: English, French, Spanish, and Arabic. The front end of the appliance is built using HTML, CSS, and JavaScript, while the back-end code relies heavily on Python.
The appliance works by first taking an audio file as input. The audio file is then converted right into a digital signal using a library called Libros. This digital signal is then passed through a neural network that has been trained on speech recognition data within the 4 languages.
The neural network consists of several layers of nodes that perform various computations on the input signal. The output of the neural network is a sequence of phonemes, that are the smallest units of sound in a language.
Once the phonemes have been identified, they’re combined into words using a language model. The language model uses statistical evaluation to find out the most certainly sequence of words that match the phonemes.
Finally, the transcribed text is displayed on the front end of the appliance using HTML, CSS, and JavaScript.
HTML
Transcription Service
Transcription Service
Transcription Result:
CSS
* {
box-sizing: border-box;
}body {
font-family: sans-serif;
margin: 0;
padding: 0;
}
.container {
max-width: 600px;
margin: 0 auto;
padding: 20px;
}
.title {
text-align: center;
font-size: 36px;
margin-bottom: 20px;
}
.form {
display: flex;
flex-direction: column;
gap: 10px;
margin-bottom: 20px;
}
.form-group {
display: flex;
flex-direction: column;
}
label {
margin-bottom: 5px;
}
input[type="file"],
select {
border: 1px solid #ccc;
padding: 10px;
border-radius: 5px;
}
.btn {
background-color: #0066CC;
color: #fff;
border: none;
padding: 10px 20px;
border-radius: 5px;
cursor: pointer;
font-size: 18px;
}
.btn:hover {
background-color: #0052A3;
}
.subtitle {
font-size: 24px;
margin-bottom: 10px;
}
.transcription {
border: 1px solid #ccc;
padding: 10px;
border-radius: 5px;
font-size: 16px;
min-height: 100px;
}
Javascript
const form = document.querySelector('#transcription-form');
const transcriptionResult = document.querySelector('#transcription-result');form.addEventListener('submit', async (e) => {
e.preventDefault();
const formData = recent FormData(form);
const response = await fetch('/api/transcribe', {
method: 'POST',
body: formData,
});
const data = await response.json();
transcriptionResult.innerText = data.transcription;
});
Python
# importing libraries
from tkinter import N
import speech_recognition as sr
import os
from pydub import AudioSegment
from pydub.silence import split_on_silence
import PySimpleGUI as sg
import os.path
from PySimpleGUI.PySimpleGUI import WIN_CLOSED
from textwrap import wrap
import smtplib
from email.message import EmailMessage
def essential():
pathForConvert = str
# First the window layout in 2 columns
# First the window layout in 2 columns
File_Choosing_Column = [
[
sg.Text("Audio Folder"),
sg.In(size=(25, 1), enable_events=True, key="-FOLDER-"),
sg.FolderBrowse(),
],
[
sg.Listbox(
values=[], enable_events=True, size=(40, 20), key="-FILE LIST-"
)
],
]
# For now will only show the name of the file that was chosen
Transcript_Viewer = [
[sg.Text("Choose an Audio file from the list on the left:")],
[sg.InputText(key='text1')],
[sg.Text(size=(40, 1), key="-TOUT-")],
[sg.Image(key="-WAVF-")],
]
# ----- Full layout -----
layout = [
[
[sg.Text("Transcript any audio (Must be in WAV format)")],
[sg.Button("Exit"), sg.Button("Run")],
[sg.Text("________________________________________________ _______________________________________________")],
sg.Column(File_Choosing_Column),
sg.VSeperator(),
sg.Column(Transcript_Viewer),
]
]
# Create the window
window = sg.Window("Transcriptor", layout)
# Create an event loop
while True:
event, values = window.read()
# Folder name was filled in, make an inventory of files within the folder
if event == "-FOLDER-":
folder = values["-FOLDER-"]
try:
# Get list of files in folder
file_list = os.listdir(folder)
except:
file_list = []fnames = [
f
for f in file_list
if os.path.isfile(os.path.join(folder, f))
and f.lower().endswith((".wav", ".mp3", ".aac", ".aiff", ".flac", ".m4a", ".ogg", ".pcm", ".wma"))
]
window["-FILE LIST-"].update(fnames)
elif event == "-FILE LIST-": # A file was chosen from the listbox
try:
filename = os.path.join(
values["-FOLDER-"], values["-FILE LIST-"][0]
)
window["-TOUT-"].update(filename)
path = filename
except:
pass
if event == "Exit":
window.close()
break
if event == "Run":
whole_text = RunIt(path)
whole_text = wrap(whole_text, 58)
eachInASeparateLine = "n".join(whole_text)
whole_text = eachInASeparateLine
window["text1"].update(whole_text)
# content
sender = "NotifAccForPy@gmail.com"
reciever = "clintfordcadio@gmail.com"
password = "ldhjjzjcffqqdizu"
msg_body = ("Here is your accomplished transcription: nn" + eachInASeparateLine)
# motion
msg = EmailMessage()
msg['subject'] = 'Text Transcription complete'
msg['from'] = sender
msg['to'] = reciever
msg.set_content(msg_body)
with smtplib.SMTP_SSL('smtp.gmail.com', 465) as smtp:
smtp.login(sender,password)
smtp.send_message(msg)
elif event == sg.WIN_CLOSED:
break
def RunIt(path):
path = path.replace("", "/", 1)
if path.endswith(".mp3"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".mp3",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".m4a"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".m4a",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".aac"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".aac",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".aiff"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".aiff",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".flac"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".flac",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".ogg"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".ogg",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
if path.endswith(".wma"):
newPathList = path.split("")
newPathString = newPathList[-1]
from os import path
import pydub
# files
src = newPathString
newPathList2 = newPathString.split("/")
newPathString2 = newPathList2[-1]
newPathString2 = newPathString2.replace(".wma",".wav",1)
del newPathList2[-1]
newPathList2.append(newPathString2)
newPathString3 = "/".join(newPathList2)
newFilenameStringWAVFinal = newPathString3
dst = newFilenameStringWAVFinal
# convert wav to mp3
sound = pydub.AudioSegment.from_file(src)
sound.export(dst, format="wav")
path = newFilenameStringWAVFinal
# create a speech recognition object
r = sr.Recognizer()
# a function that splits the audio file into chunks
# and applies speech recognition
# Splitting the big audio file into chunks
# and apply speech recognition on each of those chunks
# open the audio file using pydub
sound = AudioSegment.from_wav(path)
# split audio sound where silence is 700 miliseconds or more and get chunks
chunks = split_on_silence(sound,
# experiment with this value on your goal audio file
min_silence_len = 500,
# adjust this per requirement
silence_thresh = sound.dBFS-14,
# keep the silence for 1 second, adjustable as well
keep_silence=500,
)
folder_name = newPathString2
# create a directory to store the audio chunks
if not os.path.isdir(folder_name):
os.mkdir(folder_name)
whole_text = ""
# process each chunk
for i, audio_chunk in enumerate(chunks, start=1):
# export audio chunk and reserve it in
# the `folder_name` directory.
chunk_filename = os.path.join(folder_name, f"chunk{i}.wav")
audio_chunk.export(chunk_filename, format="wav")
# recognize the chunk
with sr.AudioFile(chunk_filename) as source:
audio_listened = r.record(source)
# try converting it to text
try:
text = r.recognize_google(audio_listened, language="fr")
except sr.UnknownValueError as e:
print("Error:", str(e))
else:
text = f"{text.capitalize()}. "
print(chunk_filename, ":", text)
whole_text += text
# return the text for all chunks detected
return whole_text
essential()
One among the important thing challenges of constructing this application was training the neural network on speech recognition data in multiple languages. I had to gather large amounts of information in each language after which use this data to coach separate neural networks for every language.
I also needed to fastidiously design the language model to be sure that it could accurately match phonemes to words in each language. This required a deep understanding of the unique characteristics of every language, akin to its grammar, syntax, and vocabulary.
Despite these challenges, we were in a position to successfully construct an ML application that may transcribe audio files into text in 4 different languages. This application has many potential use cases, akin to transcribing interviews, lectures, and meetings.
In conclusion, the usage of machine learning algorithms to automate speech recognition is an exciting field with many promising applications. Our application, which may transcribe audio files into text in English, French, Spanish, and Arabic, is only one example of the potential of this technology. As more data becomes available and our understanding of speech recognition improves, we will expect to see much more advanced applications in the long run.