Robonix Webots CI Report

Provider: deepseek · Model: deepseek-chat · Generated On: 2026-08-23T01:14:17+00:00 · Stage: success · Confidence: high

The tested commit merged PR #230 'fix(speech): split long Tencent TTS input' into dev. The change set modifies the speech service to split long Tencent TTS text into provider-safe segments, adds configurable character limits, enforces PCM-only output, and adds cancellation-aware synthesis. The Webots CI suite ran 16 scenarios covering navigation, memory, speech, and builtin capabilities; all 16 passed with no failures.

PR Changes

• Modified services/speech/README.md to document new TENCENT_TTS_MAX_CHARS and TENCENT_TTS_MAX_TOTAL_CHARS options and PCM-only contract
• Updated services/speech/docs/speech_service.proto.ref to describe pcm_s16le output instead of MP3
• Updated services/speech/package_manifest.yaml to require sample_rate=16000 and codec=pcm, adding max_chars and max_total_chars settings
• Modified services/speech/speech_service/service.py to add TTS cache bounds, validate output metadata, handle TTSInputError, and make speak() async with cancellation support
• Modified services/speech/speech_service/tencent_cloud.py to add text splitting logic, segment-based synthesis, PCM validation, and request limits
• Added services/speech/speech_service/tts_errors.py defining TTSInputError
• Added services/speech/speech_service/tts_runtime.py for request liveness tracking
• Modified services/speech/tests/test_speak_queue.py to add tests for long-text speak, short-text behavior, and cancellation
• Modified services/speech/tests/test_tencent_cloud.py to add extensive tests for text splitting, segment synthesis, limits, and cancellation
• Added services/speech/tests/test_tts_servicers.py testing one-shot and streaming TTS servicers

Test Result

• Webots CI ran 16 scenarios (flow, builtin, and cap families) covering camera/lidar snapshots, map save, memory operations, navigation, speech/speak, voiceprint list, explore, executor builtins, and scene object listing. All 16 scenarios passed (rate 1.0). The speech_speak scenario dispatched robonix/service/speech/speak and observed success with output 'spoke 35 chars on audio_driver'. The object_navigation flow also passed, navigating to a potted plant approach pose with state SUCCEEDED.

Suggested Fix

• No fix required. Follow-up: monitor the new Tencent TTS text-splitting logic in production, especially the hard-split fallback and the 40-request safety limit, to ensure long utterances are handled as expected.

Risks Or Watchouts

• The new TTS splitting logic and limits (max 40 requests, 5000 total chars) may reject very long or punctuation-dense inputs that previously worked; verify against real Tencent TTS behavior.
• The speech service now requires TENCENT_TTS_SAMPLE_RATE=16000 and TENCENT_TTS_CODEC=pcm; deployments with other settings will fail at startup.
• The scene service logged repeated 'scene-graph-llm request failed: JSONDecodeError' warnings during the run; these did not affect the passing scenarios but may indicate a separate issue with the fake VLM response format.
• The nav2 planner logged 'Failed to create a plan from potential when a legal potential was found' errors during exploration; these were expected/ignored in the passing explore scenarios but could indicate planner instability in larger maps.

Evidence