The role of emotional valence in voluntary language switching and reaction time

Project for lab rotation Cognition and Knowledge at RPTU

1. Data preprocessing

1.1 Reaction time

Chronset RT

Chronset was used to obtain the RT of each audio file. The RT results were obtained and stored as .txt for each participant, e.g., chronset_1.txt

Mannual RT

RT is accessed mannually using Audacity for a few trails in mixed block for all the participants. The mannual results were compared with the Chronset results to ensure the quality of Chronset RT.

1.2 Content of response

The content of the audio files are first analyzed using Python (speech_rec.py and speech_rec_whisper.py).

• speech_rec.py: using speech_recognition in Python to produce output files english_text.csv and can be adapted to get german_text.csv by changing the output file name and language:

  text = recognizer.recognize_google(audio, language="de-DE")

  The expected running time is around 3 hours for each language.

• speech_rec_whisper.py: using whisper in Python to produce output file transcriptions.csv, and the expected running time is about 50 hours.

Note: these two methods worked in a complementory way with the first one being more efficient and accurate. Of course, not all the audio files are accurately recognized. So, a mannual appraoch was adopted afterwards for the inaccurate ones.

1.3 Organizing data

vls_preproc.R was used to combine subject files, chronset files, and speech output, as well as rescale rating and obtain type of response etc.

• Combine all subject files
• Merge the RT from chronset files to main dataframe
• Rename and re-scale the participant rating responses
• Merge speech recognization outputs to main df
• Combine manually retrieved data
• Obtain response accuracy, language, and type.
• Add standard valence scale and compare the response valence rating to the standard scale
  • One-sample t test was used to compare the valence between paticipant response and standrd scale.
  • No significant differences were identified.
• Compare chronsst RT with manual RT in Mixed blocks
  • Normality was always violated, so Wilcoxon signed-rank test was used to compared Chronset RT and manual RT.
  • Significant differences were found between Chronst and manual RTs.

2. Data analysis

Details of the statistical analysis can be found in the file vls_analysis.R.