Audio tests related changes for kokoro and qwen3-asr

[porlows1]

🛠 Summary

Related to validation framework PR: https://github.com/intel-innersource/frameworks.ai.openvino.model-server.tests/pull/2790

https://jira.devtools.intel.com/browse/CVS-181965 Enable validation of Kokoro
https://jira.devtools.intel.com/browse/CVS-189463 voice field is not a string
https://jira.devtools.intel.com/browse/CVS-189724 Support Qwen3-ASR in OVM

🧪 Checklist

• Unit tests added.
• The documentation updated.
• Change follows security best practices.
  ``

[Copilot]

🟡 Changes recommended

_analyze_audio() still computes RMS on empty input (producing warnings/NaN), which can make the new empty-audio validation noisy or flaky.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR updates the functional test framework around OpenAI-style audio endpoints to better support Kokoro TTS and Qwen3-ASR, including request parameter handling and stronger audio-output validation.

Changes:

• Extend OpenAI audio speech request params (voice + extra body) and adjust client-side voice passing behavior.
• Improve pytest parametrization to support tests requiring both a primary and auxiliary model on the same device.
• Add word-timestamp enablement plumbing for ASR nodes and harden audio validation for empty/invalid audio responses.

File summaries

File	Description
tests/functional/utils/inference/serving/openai.py	Adds audio speech request params fields and adjusts transcription defaults.
tests/functional/utils/hooks.py	Refactors model parametrization to support auxiliary model selection per-device.
tests/functional/utils/generative_ai/validation_utils.py	Hardens audio metric computation and adds explicit empty-audio assertions.
tests/functional/utils/generative_ai/utils.py	Threads enable_word_timestamps through model/resource preparation.
tests/functional/object_model/python_custom_nodes/python_custom_nodes.py	Propagates enable_word_timestamps into calculator construction.
tests/functional/object_model/mediapipe_calculators.py	Emits enable_word_timestamps option in S2T node pbtxt.
tests/functional/object_model/inference_helpers.py	Uses OpenAI client sentinel handling for optional voice in TTS requests.
tests/functional/constants/ovms_messages.py	Updates expected voice-not-available message substring.
demos/common/export_models/export_model.py	Adds extra quantization CLI params to speech2text export path.

Review details

• Files reviewed: 9/9 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pgladkows]

I guess we could use f-string type f"{}"

[porlows1]

I agree it would look better with f-strings. For now whole file uses older .format notation and requires refactor - I propose to move it to separate task.

[mzegla]

Let's keep it with .format(...) for now to stay in line with other parts.
See line 516-517:

optimum_command = "optimum-cli export openvino --model {} --task text-to-audio --weight-format {} {} --trust-remote-code {}".format(
                source_model, precision, task_parameters['extra_quantization_params'], destination_path)

and many other places. If we switch to f-strings, we should do it everywhere across this script.

[Copilot]

🟡 Changes recommended

Audio parameters are dropped by the standard builder path, and streamed ASR responses fail when full-response mode is requested.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 10/10 changed files
• Comments generated: 2
• Review effort level: Balanced