Vowel Space Analysis
for Phonetic Research
A browser-based formant extraction and visualization platform enabling cross-linguistic vowel space comparison through Linear Predictive Coding analysis and IPA-standardized acoustic mapping.
Acoustic Phonetic Analysis in the Browser
Audioforge provides a comprehensive toolkit for vowel space visualization through extraction of the first three formant frequencies (F1, F2, F3) using Linear Predictive Coding (LPC) algorithms. The system enables researchers, clinicians, and language learners to map their vocalic productions against standardized reference data.
The tool implements the cardinal vowel system as established by Daniel Jones, projecting formant measurements onto a two-dimensional acoustic space where F1 correlates inversely with vowel height and F2 corresponds to the front-back dimension of tongue position.
With support for 41 languages across multiple language families, Audioforge facilitates cross-linguistic phonetic comparison, second language acquisition research, and clinical speech assessment with publication-quality output.
Comprehensive Phonetic Toolkit
Four integrated operating modes designed for acoustic analysis, from basic formant extraction to interactive pronunciation training with real-time articulatory feedback.
Formant Extraction
LPC-based analysis yielding F1, F2, and F3 values with configurable window length and pre-emphasis coefficients for optimal spectral resolution.
IPA Vowel Charts
Traditional trapezoid visualization with Bark-scale or Hertz plotting, conforming to IPA conventions for vowel quadrilateral representation.
Spectral Analysis
Waveform display, spectrogram generation, and MFCC extraction for comprehensive acoustic-phonetic analysis beyond formant frequencies.
Reference Database
301 vowel phonemes across 41 languages with mean formant values derived from published acoustic studies for comparative analysis.
Accuracy Scoring
Euclidean distance calculation between produced and target formants, providing quantitative assessment of pronunciation accuracy.
Export Options
Publication-quality PNG/PDF output at 300 DPI, CSV data export for statistical analysis, and JSON session files for reproducibility.
From Thesis to Tool
Audioforge emerged from empirical research conducted during my Master's thesis, "Towards Accounting for L2 Accent: The Case of Turkish Vowel Space", which investigated the phonetic transfer and acoustic realization of the Turkish vowel inventory by L1 English learners. The study distinguished between the acquisition of phonological competence (the abstract grammatical system accessible via Universal Grammar) and phonetic performance (motor-sensory execution), exploring how these factors contribute to the phenomenon of foreign accent.
The empirical analysis comprised 10 subjects (5 native Turkish speakers, 5 L1 English learners) producing 37 Turkish words anchoring the eight phonemic vowels. Formant measurements (F1, F2) were extracted using Praat at the 1/3 and 2/3 temporal points to capture the stable portion of each vowel. Key findings revealed that L1 English speakers produced Turkish vowels with more centralized formant values than their native English inventory, yet still deviated systematically from native Turkish targets—particularly in the realization of /a/ (more closed) and /i/ (more peripheral F2).
"The 'foreign accent' in L2 Turkish is primarily a phonetic rather than phonological deficiency. Because Turkish possesses a perfectly symmetrical eight-vowel inventory where each contrast (frontness, height, rounding) serves a specific phonological role in vowel harmony, the observed acoustic deviance must be attributed to the motor-movement habits transferred from the L1."
The impetus for developing Audioforge extends beyond academic phonetics. As voice-activated interfaces proliferate across consumer devices, the persistent challenge of accent-robust automatic speech recognition remains unsolved. Current ASR systems exhibit measurable performance degradation for speakers of non-standard varieties, creating accessibility barriers in human-computer interaction.
This tool aims to contribute to the broader effort of improving Large Language Model comprehension of accented speech. By providing accessible formant analysis and cross-linguistic vowel reference data, Audioforge supports the development of more inclusive speech-to-text systems that can accommodate the natural phonetic variation inherent in global speech. The goal is straightforward: enable speakers with non-native accents to interact with home devices, dictation software, and voice assistants with the same accuracy afforded to native speakers.
Cross-Linguistic Reference Data
Comprehensive vowel inventories spanning major language families, enabling typological comparison and L2 acquisition research across diverse phonological systems.
Technical Architecture
Client-side audio processing with Python backend for computationally intensive formant extraction, ensuring data privacy while maintaining analytical precision.
Begin Your Analysis
Audioforge is open source and free to use. Clone the repository to run locally or contribute to development.