Offline Cebuano Text to Speech in QnVoice for iOS with VoxCPM2
QnVoice adds offline Cebuano text to speech on iOS using a quantized VoxCPM2 runtime and a synthetic voice reference. Typed Cebuano goes directly to synthesis. English input can pass through a bundled Marian translation model. We screen five reference configurations on 12 prompts and compare two references on 16 separate prompts. The selected shared reference yielded 3.32% character error and 18.89% word error using automated Cebuano transcription. On an iPhone 15 Pro, six typed Cebuano requests reached playback in 3.14–4.67 seconds. The paper describes the iOS implementation, evaluation methods, and audio samples. Fluent-listener ratings and coverage across devices remain open.
Authors
- QnVoice
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-07
- DOI
- https://doi.org/10.5281/zenodo.24032368
- Citations
- 1
- Primary Topic
- Speech Recognition and Synthesis
- Type
- preprint