voice_audio_multi/tests/personaplex/personaplex_warm_bench_cases.json

{
  "session_core_paths": {
    "use_all_requests_by_default": true,
    "description": "PersonaPlex Python reference coverage for packaged voice prompts, assistant/service prompts, streaming user-audio encode, AR response generation, and Mimi decode.",
    "requests": [
      {
        "id": "assistant_basic",
        "input_wav": "reference/PersonaPlex/assets/test/input_assistant.wav",
        "text_prompt": "You are a wise and friendly teacher. Answer questions or provide advice in a clear and engaging way.",
        "voice_prompt": "NATF2.pt",
        "seed": 42424242,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "service_role_prompt",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You work for SwiftPlex Appliances which is a appliance repair company and your name is Farhod Toshmatov. Information: The dishwasher model is out of stock for replacement parts; we can use an alternative part with a 3-day delay. Labor cost remains $60 per hour.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424243,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "assistant_interruption_overlap",
        "input_wav": "reference/PersonaPlex/assets/test/input_assistant.wav",
        "text_prompt": "You are a concise assistant. Continue listening while responding, and recover naturally if the user talks over you.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424244,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      }
    ]
  },
  "raw_voice_prompt_wav": {
    "description": "Covers raw reference WAV voice-prompt encoding instead of loading precomputed .pt prompt embeddings. Run with --voice-prompt-dir assets/resources.",
    "requests": [
      {
        "id": "raw_voice_prompt_wav",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You are a calm support specialist. Answer briefly and keep the caller reassured while giving the next practical step.",
        "voice_prompt": "a.wav",
        "seed": 42424245,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      }
    ]
  },
  "decode_options": {
    "description": "Covers deterministic greedy decode option separately from sampled decode.",
    "requests": [
      {
        "id": "service_greedy_decode",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You work for SwiftPlex Appliances which is a appliance repair company and your name is Farhod Toshmatov. Information: The dishwasher model is out of stock for replacement parts; we can use an alternative part with a 3-day delay. Labor cost remains $60 per hour.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424246,
        "temp_audio": 0.0,
        "temp_text": 0.0,
        "topk_audio": 1,
        "topk_text": 1,
        "greedy": true
      }
    ]
  },
  "streaming_service_quality": {
    "use_all_requests_by_default": true,
    "description": "Single PersonaPlex streaming request for voice-quality regression checks. This exercises C++ streaming session state and decode_streaming with a realistic service prompt.",
    "requests": [
      {
        "id": "streaming_service_quality",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You work for SwiftPlex Appliances which is a appliance repair company and your name is Farhod Toshmatov. Information: The dishwasher model is out of stock for replacement parts; we can use an alternative part with a 3-day delay. Labor cost remains $60 per hour.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424243,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      }
    ]
  },
  "all_paths": {
    "description": "PersonaPlex full warmbench coverage in one long-lived session: packaged assistant, packaged service, overlap prompt, raw WAV voice prompt, and greedy decode.",
    "requests": [
      {
        "id": "assistant_basic",
        "input_wav": "reference/PersonaPlex/assets/test/input_assistant.wav",
        "text_prompt": "You are a wise and friendly teacher. Answer questions or provide advice in a clear and engaging way.",
        "voice_prompt": "NATF2.pt",
        "seed": 42424242,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "service_role_prompt",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You work for SwiftPlex Appliances which is a appliance repair company and your name is Farhod Toshmatov. Information: The dishwasher model is out of stock for replacement parts; we can use an alternative part with a 3-day delay. Labor cost remains $60 per hour.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424243,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "assistant_interruption_overlap",
        "input_wav": "reference/PersonaPlex/assets/test/input_assistant.wav",
        "text_prompt": "You are a concise assistant. Continue listening while responding, and recover naturally if the user talks over you.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424244,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "raw_voice_prompt_wav",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You are a calm support specialist. Answer briefly and keep the caller reassured while giving the next practical step.",
        "voice_prompt": "assets/resources/a.wav",
        "seed": 42424245,
        "temp_audio": 0.8,
        "temp_text": 0.7,
        "topk_audio": 250,
        "topk_text": 25,
        "greedy": false
      },
      {
        "id": "service_greedy_decode",
        "input_wav": "reference/PersonaPlex/assets/test/input_service.wav",
        "text_prompt": "You work for SwiftPlex Appliances which is a appliance repair company and your name is Farhod Toshmatov. Information: The dishwasher model is out of stock for replacement parts; we can use an alternative part with a 3-day delay. Labor cost remains $60 per hour.",
        "voice_prompt": "NATM1.pt",
        "seed": 42424246,
        "temp_audio": 0.0,
        "temp_text": 0.0,
        "topk_audio": 1,
        "topk_text": 1,
        "greedy": true
      }
    ]
  }
}