The order of numbers reflects the order in the usual pop-up dialogue box. Modify any settings that you wish to in the “To TextGrid (silences)” line of the script. On the other hand, I use the ELAN silence recogniser if I’m just working in a single file, as I don’t have to save or export the results of the parse and I can adjust the settings more dynamically. I use the batch-grid script when I want to process many audio files at once, and then I can import the TextGrid when I’m working on the relevant file in ELAN. The two different workflows that I have described here both have their own place in my workflows. It is also possible to train the silence recogniser what the silence level should be by selecting examples throughout the text, which is an advantage this method has over Praat (although I’ve never done this). The silence recogniser can effectively carry out the same process as that described by Eri without needing to import any files, as the settings ELAN asks for – silence level, minimal silence duration, minimal non-silence duration – are the exact same that you need to alter in Praat. However, when I tested out the recogniser, I found myself appreciating the relative simplicity of the tool. Personally, I had never really played around with the Recognizer menu before six months ago, when I started to think about pauses more deeply. This script saved me a lot of time in the field, as I could quickly process a text I’d just recorded for initial segmentation on the spot and get to transcribing within 5 minutes! It’s also helped me quickly generate segmentation for a number of pre-existing corpora for analysis purposes.ĭespite the usefulness of this script, I found myself intrigued by the built-in Silence Recognizer in ELAN. The script prompts a user to input a directory and an optional specifier for files to be processed and batch generates TextGrids which have gone through the silence recogniser. My script removes the annotation step and changes the function into To TextGrid (silences), as per Eri’s method. This script was adapted from a script by Katherine Crosswhite, which creates TextGrids automatically for annotation. If you are after clause-like units rather than intonation phrase-like units, you may not want to do this, but it’s generally recommended to do your segmentation in intonation units (Himmelmann 2006).Īnother tweak that I made is the creation of a Praat script to automate this process even further. Pauses of a length between 100ms and 250ms still play a role in the organisation of speech, although you will have to be careful of annotations which correspond to phonetic closure, rather than ‘meaningful’ pauses, and adjust your segmentation in the next pass accordingly. After reading up on pause duration (see a presentation I did with Laura Becker on this topic here), I adjusted the minimum silent interval down to 0.1 secs. This method is incredibly useful and has saved me a lot of time. This outputs a TextGrid, which can then be imported into ELAN and adjusted accordingly. In her blog post, Eri describes how to use the silence recogniser in Praat to do a first-pass segmentation of audio files. One method for reducing this bottleneck that I have made use of for the past 5 years is outlined by Eri Kashima here. In this post, I’ll describe two such methods – batch-creating TextGrids with the help of a Praat script, and silence recognition in ELAN. to 150:1! To reduce this bottleneck, you have to be creative with data processing and make liberal use of computer-assisted methods. This refers to the time sink it requires to put this data into a written format for further analysis, which is commonly estimated to be a 10:1 ratio (a native speaker requiring 10 minutes to fully transcribe and analyse 1 minute of data), but this can balloon out even further, e.g. Any linguist who works with audiovisual data is familiar with the idea of the ‘transcription bottleneck’.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |