Here are the basic steps to align recordings using the online version of UPenn’s forced aligner program (comment with questions and I will update and elaborate as needed). Nick will be putting up instructions for aligning on the lab computer.
1. Divide recording into small, managable portions of about 3 to 5 minutes long each. I usually do this with Audacity. Simply highlight subsections of the main recording, then go File > Export Selection, and choose a transparent file name such as WEP-RCT-001a (then RCT-001b, etc.).
To highlight the next portion starting exactly from the first, hover your cursor over the beginning of the previous selection, where it will turn into a hand (see Figure 1).
Figure 1 – Selecting the first subsection (cursors diappeared when taking screen shot)
If you select the beginning of the selection and move it, the end of the selection will stay anchored at that point, turning into the the start of the next selection point (see Figure 2). Then simply do File > Export Selection again and save the new file. Repeat until the entire file has been portioned up and saved in a folder specific to that speaker.
Figure 2 – Selecting the second subsection by swinging the first part around and making it the end of the second section.
2. Make an Aligner-specific transcription of each portion. This includes everything that is said, as well as codes for non-lexical noises: {SP} = small pause; {NS} = Noise; {BR} = breathing}; etc. (In a pinch, Noise will usually work for anything, including breathing, stuttering, and so forth, but be as specific as possible). Small pauses should be fairly often as long as there is the slightest pause between words. This will help to make the Alignment as close as possible.
3. Once there is a .txt (or .doc but not .docx) for every .wav file, you are ready to send the files to Aligner: http://martinet.sas.upenn.edu/PPLClient/
4. This is the easiest part, first upload the transcript files (e.g. WELS-RCT-001a.txt), then the corresponding .wav files (e.g. WELS-RCT-001a.wav), and fill in the rest of the information on the form.
5. If the transcriptions are accurate and the files are around 3 minutes, like the Arthur recordings, you should receive and email within 15 to 20 minutes. The email will be from pplaligner@sas.upenn.edu, and the attachment will be Result.TextGrid. Save and change name immediately! Otherwise you will end up with multiple “Result.TextGrid” files, named the same but different in content.
6. At this point you are ready to start correcting the textgrid, and can move on to the “How To” for correcting a textgrid.