The sample chapter rendered in each language Qwen3-TTS produces, spoken by
the same characters as the English book. Generated with the
1.7B model, the one the English audiobook was made with.
{% if not samples %}
No samples found for {{ project_name }}.
Render them with
python scripts/render_lang_test.py --project {{ project_slug }}
{% else %}
{{ samples|length }} playable
{% for s in samples %}
{{ s.code }}
{% if s.code == 'ALL' %}
All languages in one filecombined
{% else %}
{{ s.code }} sample chapter
{% endif %}
{% if s.duration %}{{ s.duration }}s · {% endif %}{{ s.size_mb }} MB
{% endfor %}
{% endif %}
Before treating these as finished: the
translations are single-pass LLM drafts, not the multi-pass pipeline in
specs/27. Grammatical and plausible, and still
open to meaning errors. These exist to judge voice fidelity and pacing per
language, not translation quality.