Optimizing Filipino Text-to-Speech Synthesis: Integration of Generative Models
Bibliographic record
Abstract
This paper explored the development of an optimized text-to-speech (TTS) system for Filipino by integrating two advanced generative models: FastSpeech 2 and Tacotron 2. Since Filipino is a low-resource language for TTS, we leveraged the Filipino Speech Corpus—which includes over 150 hours of annotated audio—to create model-specific datasets. Preprocessing steps using the Montreal Forced Aligner, phoneme-level pitch and energy normalization were implemented in FastSpeech2 to ensure precise spectral mapping. In contrast, Tacotron 2 was trained on gender-specific datasets to better capture natural prosody variations. Each synthesis pipeline paired its model with a state- of-the-art vocoder: WaveGlow for Tacotron 2 and a universal HiFi-GAN for FastSpeech 2.Objective evaluation was performed using the mel-cepstral distance metric to compare the spectral characteristics of the synthesized speech against ground truth recordings. Subjective listening tests including Mean Opinion Score (MOS) assessments and AB tests were conducted in a controlled lab environment in which the insights were provided to analyze perceptual naturalness and intelligibility of synthesized speech. Overall, the results indicate that while both models contribute to improving Filipino TTS, the FastSpeech 2 surpassed Tacotron 2 in MCD however Tacotron 2 produced higher MOS ratings and more natural-sounding speech than FastSpeech 2.This study offers a useful framework and insights in enhancing TTS systems despite using low-resource languages, which could help expand access to digital voice applications and guide future research in speech synthesis.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".