Introduction
This tutorial aims to provide a short guide on how to transcribe a Spanish audio document using a pre-installed transcription system and tLtranscribe. Differently from other TLK tools, tLtranscribe does not require any kind of configuration, which makes it ideal for users that are non-experts in ASR.
System installation
First of all, install TLK (you will find instructions in TLK installation). Then, install the TLK system for Spanish. The installation can be performed in three different ways:
-
In UNIX-compatible OSs: Download the TLK es system with dnn tarball and follow the instructions in the INSTALL file. Please note that if you do not have DNN support (CUDA), you have to download this tarball instead: TLK es system tarball.
-
In OSs compatible with .deb package files: Download the TLK es system with dnn .deb file and install it using dpkg -i tlk-system-es_1.1.0-1_all.deb. Please note that currently this transcription system has only been tested in Ubuntu (trusty) and Debian (wheezy). Please note that a CUDA-compatible GPU is required as this transcription system uses Deep Neural Networks (DNN). If you do not have a GPU with CUDA support, please download this package instead: TLK es system .deb file.
-
In Ubuntu or Debian: Add the TLK Ubuntu or Debian repository in order to install it using APT.
-
First add the TLK public key to your APT keys:
$ wget -O - www.translectures.eu/tlk/repos/apt/translectures-tlk@dsic.upv.es.gpg.key | sudo apt-key add -
-
Then edit your “/etc/apt/sources.list” file, adding the following lines:
deb http://www.translectures.eu/tlk/repos/apt/(debian|ubuntu) DISTRO main deb-src http://www.translectures.eu/tlk/repos/apt/(debian|ubuntu) DISTRO main
where “DISTRO” is the name of your Ubuntu or Debian distribution, e.g. “trusty”.
-
Now you can install the TLK transcription system package by doing
$ sudo apt-get update sudo apt-get install tlk-system-es
or in the case of Ubuntu trusty, in which there is DNN support:
$ sudo apt-get update sudo apt-get install tlk-system-es-dnn
-
Using tLtranscribe
You can download an example lecture (test video), or use one of your own. Then simply execute:
$ tLtranscribe test-video.mp4
tLtranscribe will automatically transcribe the video using the tLtask-segment and tLtask-recognise tools. Note that the execution of this tool is time-consuming and it requires a considerable amount of RAM memory. By default, all available cores will be used, and each of them requires at least 4GB of RAM memory. Multi-core execution can be disabled by using the option --only-one-core.
Upon a successful execution, subtitles will be saved in the file video-file.srt; in the case of our test video, that would be test-video.mp4.srt. Finally, the subtitles can be viewed using your favourite multimedia player, e.g.,
$ mplayer test-video.mp4 -sub test-video.mp4.srt