2.2 KiB
2.2 KiB
| title | description | tags | source | created | updated | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| google-cloud-texttospeech: model_name Silently Ignored Below v2.28.0 | VoiceSelectionParams.model_name field is silently ignored in google-cloud-texttospeech < 2.28.0; requires ≥ 2.36.0 for reliable behavior |
|
daily/2026-05-08.md | 2026-05-08 | 2026-05-08 |
google-cloud-texttospeech: model_name Silently Ignored Below v2.28.0
The Problem
VoiceSelectionParams.model_name (used to select Gemini/Journey voices) is silently ignored in google-cloud-texttospeech versions below 2.28.0. The TTS call succeeds, returns audio — but falls back to a default voice with no error or warning. This makes it look like the voice selection is working when it isn't.
voice = texttospeech.VoiceSelectionParams(
language_code="en-US",
model_name="en-US-Journey-F", # silently ignored if lib < 2.28.0
)
Version Requirements
| Version | Behavior |
|---|---|
| < 2.28.0 | model_name silently ignored — default voice used |
| 2.28.0–2.35.x | model_name accepted but may have edge-case bugs |
| ≥ 2.36.0 | Fully reliable; recommended minimum |
Fix
In pyproject.toml:
[tool.poetry.dependencies]
google-cloud-texttospeech = ">=2.36.0"
After updating:
poetry lock --no-update # regenerate lock without upgrading everything else
# Then rebuild the Docker container
docker compose build worker-ai && docker compose up -d worker-ai
Why It Happens
The model_name field was added to the proto definition in a later release. Older versions of the generated client simply drop unknown fields when serializing the request — no exception, no warning logged by the client library.
Diagnostic Signal
If TTS produces audio in a different voice than specified (especially a generic/robotic voice instead of a Journey voice), check the library version first:
pip show google-cloud-texttospeech | grep Version
Related
- wiki/concepts/google-cloud-tts-vs-ai-studio-tts — full Cloud TTS vs AI Studio comparison
- wiki/concepts/poetry-docker-version-mismatch — other silent library version issues