In addition to direct voice input, Speech Pad can extract speech from HTML5 audio and video files and YouTube clips. This can be achieved either by placing the microphone near the speakers or using audio cable (physical or virtual), or using stereo mixer.
Transcription of audio and video files with the length more than 15 minutes are the extended features of SpeechPad. For use them you must turn on extented features in the user profile page (small fee will be charged).
Press the +transcription button in Speechpad.pw to open the transcription module and get started.
The Speechpad.pw transcription module
To begin transcribing, load audio or video to the player and put your microphone near the speakers, then press the Start recording button.
The Noise protection drop down list is used for poor quality and noisy audio and prevents jam of voice input.
The Length of preview buffer setting prevents to accumulate too much symbols in this buffer, this happen if the speaker does not make pause in his speech.
There are two mode of transcribing: automatic and semiautomatic. In automatic mode the checkbox Run synchronously with the recording is checked.
Transcribing algorithm in automatic mode is as follows:
1) load an audio/video file or video clip into the player
2) the sound from the player is sent to the microphone using stereo mixer or virtual audio cable
3) set transcription module settings, check the Insert time labels checkbox
4) press start recording button
If Run synchronously with the recording checkbox is not checked, the fields Play time, Pause time and the Play button will be shown, as in the screenshot below.
This mode is for making “semiautomatic” audio to text translation, with the user as an intermediary. Listen to the audio through headphones and then speak the sentences you hear into the microphone during the pauses. The values of play time and pause time can be adjusted to maintain a comfortable pace for the re-translation. If the values are not set then you can press play/stop button to make pauses.
Word processing after transcription
The text obtained by speech recognition contains errors. To correct them can use the time stamps obtained when transcribed. In this mode, you must also disable the check box run synchronously with the recording or use hotkey to start / stop the player (hotkey is available if SpeechPad extension is installed).
Correction algorithm is as follows:
1) normalize the text with the time stamps by pressing time labels to SRT an then SRT to time labels
2) check the checkbox start from time labels
3) uncheck the checkbox Run synchronously with the recording
4) position the cursor at the desired location in the text
5) by using a hot key or by pressing play/pause button for the player listen to the corresponding piece of the audio file. The player starts from the left nearest time stamp
6) manually edited all text fragments
7) at the end of correction remove the time stamps (remove time labels button) or translate text in Youtube format (time labels to SRT button).