Generate multilingual 24 kHz speech in Modly with the built-in CustomVoice speakers from Qwen3-TTS 1.7B.
An open-source Python PROCESS extension by DrHepa that maps one Modly text input to one mono PCM16 WAV file at 24 kHz using the built-in speakers from Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice.
Version 0.1.6 supports the unmodified Modly v0.4.1 Python PROCESS setup and runtime contracts, including their payloads without a model-directory field. Its Version 0.1.6 error-routing changes have static, unit, and mocked protocol validation; prepared-runtime and packaged E2E remain pending. The limited real-runtime evidence predates the Version 0.1.6 source changes. In one clean Linux runtime, setup completed, the immutable 4,520,218,951-byte model inventory was fully size/hash verified, and one real scalar PROCESS request loaded the pinned snapshot strictly offline and produced a structurally valid mono PCM16 WAV at 24 kHz.
Production Install-from-GitHub evidence also showed that multi-gigabyte Python dependency downloads can remain active for many hours on a healthy connection. Dependency-install commands therefore have no total elapsed-time timeout; they remain foreground operations and retain bounded private output capture without changing dependency or model pins.
That evidence applies only to the tested Linux runtime configuration. It is not a perceptual or transcription-quality evaluation, and it does not establish Windows, Linux x64, Linux ARM64, CPU, CUDA 12.8, or CUDA 13.0 support as a whole. Clean Install-from-GitHub UI validation and packaged Windows/Linux E2E remain pending. The source-only GitHub Actions matrix has completed successfully on Ubuntu and Windows with Python 3.11 and 3.12; that is not packaged native or runtime E2E evidence.
Read the full README on GitHub →