Taiwan Mandarin Nasal Coda(in/jing) Project

대만 민남어 화자의 국어 음절말 비음 실현 변이

Overview

This repository documents my research assistant (RA) work on an acoustic-phonetic study of nasal coda realization and merger in Taiwan Mandarin, focusing on syllables of the form in and jing.
The goal of the project is to investigate whether alveolar and velar nasal codas (/n/ vs. /ŋ/) are acoustically distinguishable despite perceptual merger.

This repository is intended as a personal research portfolio, demonstrating my experience with perceptual judgment, Praat-based segmentation, annotation, and acoustic analysis.

---

Contents

.
├── README.md
├── scripts/
│   ├── analysis_in_2nd.praat
│   └── analysis_jing_2nd.praat
├── data/
│   ├── raw/
│   │   ├── wav_in/
│   │   ├── wav_jing/
│   │   ├── textgrid_in/
│   │   └── textgrid_jing/
│   └── derived/
│       ├── result_in.csv
│       └── result_jing.csv
└── docs/
    └── variable_dictionary.md

---

My Role as a Research Assistant

In this project, I was responsible for the following tasks:

1. Perceptual judgment

   • Listening to original recordings and judging whether in and jing tokens were perceived as /in/ vs. /iŋ/ (e.g. “인/잉”, “찐/찡”)

2. Segmentation and extraction

   • Identifying and extracting target syllables (in, jing) from original recordings
   • Classifying tokens according to speaking context:
     • sentence-medial (mid)
     • sentence-final (fin)
     • word-reading task, first repetition (w1)
     • word-reading task, second repetition (w2)

3. Manual annotation

   • Creating TextGrid files in Praat
   • Segmenting each token into vowel and nasal portions

4. Acoustic analysis

   • Measuring vowel and nasal segments separately
   • Extracting acoustic features at 0%, 25%, 50%, 75%, and 100% of each segment:
     • Formant values (F1, F2, F3)
     • Spectral measures related to nasality (A1, P1)
   • Automating extraction using Praat scripts

---

File Naming Convention

File names encode speaker identity, material type, and speaking context as follows:

[SpeakerID]_[Material]_[Target]_[Context].wav

Components

• SpeakerID: two-digit number (01–06) identifying the speaker
• Material:
  • A: sentence-reading material (Group A)
  • W: word-reading task
• Target:
  • in: 銀
  • jing: 京
• Context:
  • mid: sentence-medial position
  • fin: sentence-final position
  • 1: first repetition in word-reading task
  • 2: second repetition in word-reading task

Examples

• 01_A_in_mid.wav
• 01_A_in_fin.wav
• 01_W_in_1.wav
• 01_W_in_2.wav

The same naming scheme applies to corresponding TextGrid files.

---

Data Description

Raw Data (data/raw/)

• wav/: original audio recordings
• textgrid/: manually annotated Praat TextGrid files
  • Vowel and nasal segments are explicitly labeled

All raw data are shared with permission and may be reused for research purposes.

Derived Data (data/derived/)

CSV files containing acoustic measurements extracted from Praat.

Key Variables

• V_Dur: duration of the vowel segment (seconds)
• N_Dur: duration of the nasal segment (seconds)
• Total_Dur: total duration of vowel + nasal
• V_Ratio: normalized vowel duration (V_Dur / Total_Dur)
• N_Ratio: normalized nasal duration (N_Dur / Total_Dur)
• i_F1_*, i_F2_*, i_F3_*: vowel formant values at 0/25/50/75/100%
• n_F1_*, n_F2_*, n_F3_*: nasal formant values at 0/25/50/75/100%
• i_A1, i_P1: spectral peaks for the vowel segment (50% point)
• n_A1, n_P1: spectral peaks for the nasal segment (50% point)

A full variable description is provided in docs/variable_dictionary.md.

---

Analysis Scripts

Praat scripts used for automated extraction:

• analysis_in_2nd(0,25,50,75,100).praat
• analysis_jing_2nd(0,25,50,75,100).praat

How to Run

1. Open Praat.
2. Modify the directory paths at the top of the script if needed.
3. Ensure .wav and .TextGrid files share the same base filename.
4. Run the script to generate CSV output files.

---

Notes

• This repository focuses on data preparation and acoustic measurement.
• Statistical analyses (e.g. speaker effects, duration ratios, formant trajectories) can be conducted using Excel, SPSS, or R based on the derived CSV files.

---

Citation

If you use materials from this repository, please cite this repository and the original study that motivated the analysis.

---

Contact

For questions about the data or analysis pipeline, feel free to contact me via GitHub.