bc-dunia/qwen3-TTS-studio v0.1.6 on GitHub

This release focuses on making the Docker image build and run reliably (fixes GitHub issue #5).

Docker / GHCR

Fixed a Dockerfile ordering bug where COPY --chown=appuser:appuser ran before appuser existed (build failure).
Stopped masking apt-get failures; switched Debian mirrors to HTTPS and added retries to reduce transient 502 errors.
Added missing runtime dependencies required by the Qwen TTS stack (ffmpeg, sox, libsndfile1, libgomp1, etc.).
Set HF_HOME for Hugging Face cache inside the container.
Updated healthcheck to GET / (works across Gradio versions).
GHCR publishing now builds multi-arch images (linux/amd64 + linux/arm64) and updates the latest tag on releases.

Improved error message when speech tokenizer weights need to be copied but the bind-mounted model directory is not writable.
Switched to dtype= when calling from_pretrained to avoid deprecation warnings.

Clarified Docker instructions: .env is optional unless you use Podcast/LLM providers.
Added an end-to-end Docker smoke test that generates a WAV on the host.
Added macOS Docker note (no MPS inside Linux containers) and memory guidance.