[voice] Add dedicated STT latency and TTS generation speed benchmarks

[Copilot]

The bench_voice_assistant.cpp had a Stubs: 1 annotation and no isolated benchmarks for STT or TTS performance — only full-pipeline (processVoiceCommand) measurements that bundle STT + LLM + TTS together, making it impossible to attribute latency regressions.

Changes

benchmarks/bench_voice_assistant.cpp

STT Latency Benchmarks — exercise STTProcessor::transcribe() directly:

• BM_STTLatency_Short — 1 s fixed baseline
• BM_STTLatency_ByDuration — 0.5 / 1 / 5 / 30 / 60 s inputs (latency-vs-duration scaling)
• BM_STTLatency_WithDiarization — speaker diarization path (1 / 5 / 30 s)
• BM_STTLatency_Streaming — streamTranscribe() callback path (1 / 5 / 30 s)
• BM_STTStatistics — getStatistics() overhead

TTS Generation Speed Benchmarks — exercise TTSProcessor::synthesize() directly:

• BM_TTSGenSpeed_Short — short phrase baseline
• BM_TTSGenSpeed_ByLength — 50 / 200 / 500 / 2000 char texts
• BM_TTSGenSpeed_WithOptions — speed variants 0.5× / 1.0× / 1.5× / 2.0×
• BM_TTSGenSpeed_Streaming — streamSynthesize() callback throughput
• BM_TTSAvailableVoices / BM_TTSStatistics — auxiliary overhead

Helpers added:

• buildWavBlob(duration_ms) — generates a standards-compliant 16-bit PCM WAV of arbitrary duration (noise, deterministic seed) for use as STT input without requiring a real audio file
• getSTTProcessor() / getTTSProcessor() — lazy-initialised singletons; initialization cost paid once per binary run, not per benchmark iteration

Each benchmark emits items_processed, bytes_processed (where relevant), and domain counters (audio_duration_ms, text_chars, speed_x10) for CI regression tracking.

File header updated: Stubs: 0, version 0.0.33, quality score 100.0.

src/voice/ROADMAP.md

Performance benchmark checklist item updated from [I] → [P].

Type of Change

• Bug fix
• New feature
• Refactoring
• Documentation
• Other:

Testing

• Unit tests added/updated
• Integration tests added/updated
• Manual testing performed

📚 Research & Knowledge (wenn applicable)

• Diese PR basiert auf wissenschaftlichen Paper(s) oder Best Practices?
  • Falls JA: Research-Dateien in /docs/research/ angelegt?
  • Falls JA: Im Modul-README unter "Wissenschaftliche Grundlagen" verlinkt?
  • Falls JA: In /docs/research/implementation_influence/ eingetragen?

Relevante Quellen:

• Paper:
• Best Practice:
• Architecture Decision:

Checklist

• Code follows project style guidelines
• Self-review completed
• Documentation updated (if needed)
• No new warnings introduced

Original prompt

---

This section details on the original issue you should resolve

<issue_title>[voice] Improve STT latency and TTS gen speed benchmarks</issue_title>
<issue_description># [voice] Improve STT latency and TTS gen speed benchmarks

Summary

• Module: voice
• Item: Performance benchmarks (STT latency, TTS generation speed)

Implementation Context

• Roadmap section: Production Readiness Checklist
• Enhancement hint: Line 14: Open Issues: TODOs: 0, Stubs: 1

Implementation Phases

Phase 1: Voice Pipeline & Session Management (Status: Completed )

• VoiceAssistant central coordinator for all voice interaction
• STT processing via Whisper AI (speaker diarization, timestamps)
• LLM integration via EmbeddedLLM / LlamaWrapper (intent recognition, query generation, response generation)
• TTS synthesis with audio format output
• Voice command processing pipeline (audio STT LLM TTS audio)
• Session state and conversation history management
  Phase 2: Streaming STT & Wake-Word Detection (Status: In Progress )
• Real-time streaming STT (word-by-word transcription as audio arrives)
• Wake-word detection for hands-free activation
• Multi-speaker diarization improvements
  Phase 3: Voice Macros & Browser Streaming (Status: Planned )
• Voice command macros (user-defined shortcuts to AQL queries)
• Language detection and automatic locale switching
• Noise suppression preprocessing (RNNoise integration)
• WebSocket audio streaming endpoint for browser clients
• Voice session playback and search in stored transcripts
  Phase 4: Multi-Language TTS & Biometric Authentication (Status: Planned )
• Multi-language TTS (German, French, Spanish voices)
• Emotion / sentiment detection from voice tone
• Voice biometric authentication (speaker verification)
• Real-time meeting transcription with action-item extraction
• Integration with telephony systems (SIP / WebRTC)

Related Work

• No direct overlap identified in pre-check.
• Relationship is for coordination only; do not expand this issue scope.

Scope Boundary

• In scope: implement only this roadmap item: "Performance benchmarks (STT latency, TTS generation speed)"
• Out of scope: any other roadmap checklist item, even from the same module/section
• Related issues/PRs are references only; implementation remains isolated to this issue
• If additional work is required, open/link a follow-up issue instead of extending this issue

Mandatory Delivery Workflow (ThemisDB Rules)

• Phase 0: Existing code review before implementation
  • Identify existing files/symbols/interfaces and document reuse plan
  • Verify no duplicate implementation of existing functionality
  • Record affected files and integration points before coding
• Scope gate (must pass before coding)
  • Confirm implementation target is only the single issue description
  • Confirm no additional roadmap items are implemented in this issue/PR
  • Any newly discovered work is split into separate linked issue(s)
• Design and implementation rules reviewed from:
  • docs/analysis/IMPLEMENTATION_GUIDE.md
  • docs/analysis/GPU_ACCELERATION_ADDENDUM.md
  • docs/architecture/MODULAR_ARCHITECTURE_ROADMAP.md
  • docs/architecture/THEMIS_CORE_GUIDE.md
  • docs\en\features\voice_assistant_guide.md
  • docs\en\voice\VOICE_MODULE_GUIDE.md
• Architecture and compatibility validation
  • Confirm behavior compatibility with existing APIs unless breaking change is explicitly declared
  • Confirm telemetry/logging/metrics integration follows existing module patterns
• Code review gate before completion
  • Self-review against roadmap acceptance criteria
  • Cross-check for overlap/duplication with existing implementations
• Validation gate
  • Unit + integration tests updated/added
  • Performance impact measured against baseline

Implementation Tasks

• VoiceAssistant central coordinator for all voice interaction
• STT processing via Whisper AI (speaker diarization, timestamps)
• LLM integration via EmbeddedLLM / LlamaWrapper (intent recognition, query generation, response generation)
• TTS synthesis with audio format output
• Voice command processing pipeline (audio STT LLM TTS audio)
• Session state and conversation history management
• Real-time streaming STT (word-by-word transcription as audio arrives)
• Wake-word detection for hands-free activation
• Multi-speaker diarization improvements
• Voice command macros (user-defined shortcuts to AQL queries)
• Language detection and automatic locale switching
• Noise suppression preprocessing (RNNoise integration)
• WebSocket audio streaming endpoint for browser clients
• Voice session playback and search in stored transcripts
• Multi-language TTS (German, French, Spanish voices)
• Emotion / sentiment detection from voice tone
• Voice biometric authentication (speak...

• Fixes [voice] Improve STT latency and TTS gen speed benchmarks #2357

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.