Offline Urdu Text to Speech for iOS with Distilled Matcha

QnVoice distills a pinned Urdu Matcha acoustic model to three flow steps with a learned mel correction and runs HiFi-GAN offline. On an iPhone 15 Pro, median warm synthesis time fell from 1.50 s to 1.30 s across 72 runs per model. The companion project contains the ONNX files, benchmark records, and paired samples. Native-speaker listening tests and commercial release review remain pending.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-07
DOI
https://doi.org/10.5281/zenodo.24029948
Primary Topic
Speech Recognition and Synthesis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Offline Urdu Text to Speech for iOS with Distilled Matcha

QnVoice
3 links
Zenodo (CERN European Organization for Nuclear Research)
Speech Recognition and Synthesis
preprint

Offline Urdu Text to Speech for iOS with Distilled Matcha

QnVoice
preprint en

Abstract

QnVoice distills a pinned Urdu Matcha acoustic model to three flow steps with a learned mel correction and runs HiFi-GAN offline. On an iPhone 15 Pro, median warm synthesis time fell from 1.50 s to 1.30 s across 72 runs per model. The companion project contains the ONNX files, benchmark records, and paired samples. Native-speaker listening tests and commercial release review remain pending.

Zenodo (CERN European Organization for Nuclear Research)
Speech Recognition and Synthesis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon