Auto Transcribing using AI-ish thing? No. Not yet.

Google docs integrated Voice Recognition. What a brilliant tool. Or is it?

It is safe to call that auto-transcribing voice data is mankind’s dream. I know it’s not possible to be come true right now, but I did some digging anyway, and found out that some people claim that using Google docs integrated Voice Recognition can be useful for transcribing.

But I was skeptical. Very, very skeptical. 

Because, well, I know voice recognition is still shitty (for Japanese language at least), and I also tried Microsoft Watson for demo transcribing in the past, and it was, well, shitty.

But well, if it can be automated, then I’m golden, amn’t I?

So I wrote transcripts for AutoHotkey to minimize my effort for clicking absolutely everywhere, and I succeeded to develop an environment.

Here is some demo. Because I wrote Ahk script, I just placed everything there, and hit Esc key. Nothing more.

As you can see, as long as they are talking clearly, (but data itself is super clear enough for transcribing manually), they can do pretty decent job.

But because I needed to connect directly from your earphone jack to microphone jack by using 3.5mm male to male audio cable, and it somehow degrades sound. It sounded crappy when I heard the screen captured sound. It sounds more decent in the actual file. One decent sound card may solve this. or not.

In any case, Google doc voice recognition part just stops working sometimes without any particular reason, and it can be annoying. You have to stare all the time anyway, so it’s more like a Auto-pilot in training.

Also, it is not good enough for people who needs serious transcribing, like Conversational Analyst, such as myself. Because you will need a tons of time to re-construct those transcript after auto-doing. You will wish that you just start typing using vtt3 or something. It is, i guess, much faster than absolutely nightmare re-construction.

So who is this for? Well, if you have really clear, slowly-speaked data without so-called backchanneling , such as interviews, it may work. Or if you want to just scrub voice data from audio, you may be able to use it.

But for me? meh, no. Not yet, at least.

MacWhinney, B. (2007) “The TalkBank Project” J. C. Beal, K. P. Corrigan & H. L. Moisl eds., Creating and Digitizing Language Corpora: Synchronic Databases, Vol.1. Houndmills: Palgrave-Macmillan.

*If you are interested in using this script, you need to change quite many pieces. Ask me if you don’t know what to do, via comment below, or twitter, or whatever.

/*
script written by neokix (@neo_kix)
website and source code: https://neokixblog.wordpress.com/

;;;Reminder;;;
Window Key-> #
Ctrl Key -> ^
Alt Key -> !
Shift Key -> +
無変換 -> {vk1D}
変換 -> {vk1C}
かなkey -> {vkF2}
IME - > {vkF3}, {vkF4}
CapsLock ->{vkF0}
F1 & Up:: -> you can use non-modifier keys as modifier keys, maybe.
*/
#NoEnv  ; Recommended for performance and compatibility with future AutoHotkey releases.
; #Warn  ; Enable warnings to assist with detecting common errors.
SendMode Input  ; Recommended for new scripts due to its superior speed and reliability.
SetWorkingDir %A_ScriptDir%  ; Ensures a consistent starting directory.
#SingleInstance, Force
SetTitleMatchMode, 2
return

Esc::
	Settimer, autopilot, 3000
	return

autopilot:
	vrecordingon =
	vrecordingoff =
	IfWinExist, ahk_class Chrome_WidgetWin_1 ;Google Chrome
	{
	WinActivate, ahk_class Chrome_WidgetWin_1 ;Google Chrome
	MouseClick, Left, 907, 285
	gosub, recordingon
	Sleep, 200
	Controlsend, , {Space}, VLC
	Sleep, 15000
	Controlsend, , {Space}, VLC
	Send, {Enter}
	gosub, recordingoff
	sleep, 100
	}	
	else,
	{
	msgbox, oh my!
	}
	return

recordingon:
	If (vrecordingon =)
		{
		while (vrecordingon =) 
			{
			PixelGetColor, vrecordingon, 907, 285
			sleep, 500
			}
		}
	else if (vrecordingon = 494949) ;gray
		While (vrecordingon = 494949)
		{
		MouseClick, Left, 907, 285
		Sleep, 300
		PixelGetColor, vrecordingon, 907, 285
		Sleep, 300
		}
	else
		{
		}
	return

recordingoff:
	If (vrecordingoff =)
		{
		while (vrecordingoff =) 
			{
			PixelGetColor, vrecordingoff, 907, 285
			sleep, 500
			}
		}
	else if (vrecordingoff = DC5034) ;red
		While (vrecordingoff = DC5034)
		{
		MouseClick, Left, 907, 285
		Sleep, 300
		PixelGetColor, vrecordingoff, 907, 285
		Sleep, 300
		}
	else if (vrecordingoff = FFFFFF) ;white
		While (vrecordingoff = FFFFFF)
		{
		MouseClick, Left, 907, 285
		Sleep, 300
		PixelGetColor, vrecordingoff, 907, 285
		Sleep, 300
		}
	else
	{
	}
	return
	
F12::
	Settimer, autopilot, off
	return
	

 

広告

コメントを残す

以下に詳細を記入するか、アイコンをクリックしてログインしてください。

WordPress.com ロゴ

WordPress.com アカウントを使ってコメントしています。 ログアウト /  変更 )

Google+ フォト

Google+ アカウントを使ってコメントしています。 ログアウト /  変更 )

Twitter 画像

Twitter アカウントを使ってコメントしています。 ログアウト /  変更 )

Facebook の写真

Facebook アカウントを使ってコメントしています。 ログアウト /  変更 )

w

%s と連携中

This site uses Akismet to reduce spam. Learn how your comment data is processed.