README.md

Generate Text From Video using Python

Install the following python modules:

pip3 install librosa SpeechRecognition

Install ffmpeg CLI app

video_to_text.py

import os
import math
import librosa # pip3 install librosa
import speech_recognition as sr # pip3 install SpeechRecognition

# install `ffmpeg`

# Goes over every singl e video in the input directory, then converts to mp3, to generate subtitles
inputDir='inputs' # input videos
subDir='subs'     # output subtitle directory
outputDir='out'   # output mp3

def makeDir(name: str) -> None:
    if not(os.path.exists(name)):
        os.mkdir(name)

makeDir(inputDir)
makeDir(subDir)
makeDir(outputDir)

for file in os.listdir(inputDir):
    inputFile = os.path.join(inputDir, file)
    # checking if it is a file
    if os.path.isfile(inputFile):
        filename = file.split('.')[0]

        print(f'- Processing "{filename}":')

        outputFile = f'{outputDir}/{filename}.wav'

        # generate audio if doesn't exist
        if not(os.path.exists(outputFile)):
            print(f'\t* [Process] Converting {filename} to wav...')
            command2wav = f'ffmpeg -i {inputFile} {outputDir}/{filename}.wav 2> /dev/null'
            os.system(command2wav)
            print(f'\t* [Done] {filename} converted to wav.')
        else:
            print(f'\t* [Skip] {filename} Audio file already exist.')

        subFile = f'{subDir}/{filename}.txt'

        if not(os.path.exists(subFile)):
            print(f'\t* [Process] Generating subtitle for {filename}...')
            r = sr.Recognizer()

            # Generate subtitle by chunk per 120 seconds
            durationPerCycle = 120
            totalDuration = math.floor(librosa.get_duration(path=outputFile))

            duration = 0
            offset = 0

            # Loop until the entire video is generated
            while duration < totalDuration:

                offset += duration
                duration += durationPerCycle

                # if 5 seconds left, skip it. Edit this value if 5 seconds fail for you
                if (totalDuration - (duration - durationPerCycle)) < 5:
                    offset = duration - durationPerCycle
                    duration = totalDuration
                    print(f'\t\t* [Fail] Unable to transcribe from {offset / 60} to {duration / 60} minutes')
                    break;

                # set the remaining duration if duration is less than per cycle
                if duration >= totalDuration:
                    offset = duration - durationPerCycle
                    duration = totalDuration

                print(f'\t\t* Transcribing from {offset / 60} to {duration / 60} out of {totalDuration / 60} minutes.')
                audio = sr.AudioFile(outputFile)
                with audio as source:
                    audio = r.record(source, duration=duration, offset=offset)
                    transcribe = r.recognize_google(audio)

                    # Append to subtitle file
                    with open(f'{subDir}/{filename}.txt', 'a') as subOutFile:
                        subOutFile.write('\n'+transcribe)
                        print(f'\t\t* [Done] Subtitle generated from {offset / 60} to {duration / 60} for {filename}.')


            print(f'\t* [Complete] {filename} is processed!\n')
        else:
            print(f'\t* [Skip] Subtitle for {filename} is already exist.\n')