fix: murf tts voice config with each tts request

[nitin4real]

No description provided.

[github-actions]

Review — fix: murf tts voice config with each tts request

Thanks for the fix. The core change (filter params into voice_config / advanced_settings, send voice config with each text chunk) looks sensible. A few items worth addressing before merge:

Correctness / API contract

• Wire-format key rename from voiceConfig → voice_config (murf_tts.py:205). This changes the payload sent to Murf's websocket. Please confirm against Murf's current API spec and call it out in the PR description — if you deploy this while Murf still expects voiceConfig, TTS will silently break. Worth a note in the PR body and, ideally, a link to Murf's docs.

• manifest.json dropped style/rate/pitch/variation from the schema, but VOICE_CONFIG_PARAM_KEYS in murf_tts.py still lists them. Result: users configuring those keys via property.json/runtime config may be rejected by the manifest validator before the code ever sees them. Either restore the manifest fields (plus pronunciation_dictionary) so configs can carry them, or drop them from VOICE_CONFIG_PARAM_KEYS too. The new test test_get_voice_config_from_params_filters_to_supported_keys passes a style in params, but that path is never exercised through real manifest validation, so the test gives a false sense of coverage.

• Dual-alias keys in VOICE_CONFIG_PARAM_KEYS (voice_id + voiceId, multiNativeLocale + multi_native_locale, pronunciationDictionary + pronunciation_dictionary). If a user sets both, both land in the payload and Murf may reject it or prefer one unpredictably. Either document which form is canonical, or normalize (e.g. snake → camel) before sending.

Performance

• _send_text_internal now attaches the full filtered voice_config to every chunk. For streaming responses with many small chunks this meaningfully inflates per-message size (especially if pronunciation_dictionary is present). Worth verifying Murf's API actually requires per-chunk voice config — or whether one-shot on connect + re-send only when it changes would suffice.

Logging

• Many log_debug → log_info promotions throughout murf_tts.py and extension.py (lifecycle, "KEYPOINT Sending voice config...", websocket connect/close, client init/close). This will noticeably increase INFO-level log volume in production. If this was for local debugging, revert before merge; if intentional, consider whether connect/close chatter truly warrants INFO.
• Inconsistent: _send_text_internal's "KEYPOINT Sending text to MURF TTS" remained log_debug while _send_voice_config's equivalent was promoted to log_info.

Tests

• tests/test_voice_config.py uses a hardcoded Path(__file__).resolve().parents[6] sys.path hack. This is fragile — any directory rename breaks it. Existing tests in this extension rely on conftest.py + normal package imports; please follow the same pattern (or add a local conftest.py rather than path manipulation).
• Tests use object.__new__(MurfTTSynthesizer) to bypass __init__. It works, but couples tests tightly to class internals; consider a lightweight constructor/factory if this pattern spreads.
• Missing coverage for: config.params is None, conflicting aliases (voice_id + voiceId both set), and an empty voice_config in _send_text_internal.

Minor / nits

• .env.example: extra blank line before # murf tts, and file still lacks a trailing newline.
• manifest.json bump 0.1.0 → 0.1.1 is fine for a bug fix. If the schema removals are intentional (a breaking config change), this arguably warrants 0.2.0 under semver.
• Blank line after class MurfTTSynthesizer: before def __init__ is unconventional — harmless, but other classes in this file don't do it.

Nice-to-have

• A brief README note about the new advanced_settings params (min_buffer_size, max_buffer_delay_in_ms) and the per-chunk voice_config behavior would help downstream users.

Overall the direction looks right — the main blockers for me are confirming the voiceConfig → voice_config wire change is correct, and reconciling the manifest schema with VOICE_CONFIG_PARAM_KEYS.