[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79H0Z6ZCSMHZGRZT4XNPKZ1":3,"skill-same:01M79H0Z6ZCSMHZGRZT4XNPKZ1":44},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":14,"categories":15,"keywords":16,"publishedAt":17,"linkOnly":18,"markdownContent":19,"markdownBody":20,"frontMatter":21,"refName":22,"commitSha":23,"skillDirectory":24,"forksCount":25,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":28,"updatedAt":29,"install":30,"resources":43},"01M79H0Z6ZCSMHZGRZT4XNPKZ1","hl-build-voice-app",null,"Datnewlevel\u002FA.I-drone-pi5-hailo-hat","Datnewlevel","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F189353576?v=4","01M79GZD54F2AF7CXVBEKNXX3R","01M79GZD5RCAPCM33SR8Y3Y220",2,"MIT","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-voice-app\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: hl-build-voice-app\ndescription: Build voice apps for **all Hailo accelerators**: Whisper STT on Hailo-8\u002F8L\u002F10H + optional Piper TTS on CPU.\n---\n\n# Skill: Build Voice Application\n\nBuild voice apps for **all Hailo accelerators**: Whisper STT on Hailo-8\u002F8L\u002F10H + optional Piper TTS on CPU.\n\n## When This Skill Is Loaded\n\n- User wants **speech input** or **speech output** in a Hailo app\n- User mentions: voice, speech, Whisper, TTS, microphone, STT, speak, listen\n- User wants to add voice to an existing LLM or VLM app\n- User wants speech recognition on Hailo-8 or Hailo-8L\n\n## Hardware Compatibility\n\n| Feature | Hailo-8\u002F8L | Hailo-10H |\n|---|---|---|\n| STT (Whisper) | ✓ via `InferModel` (encoder+decoder HEFs) | ✓ via `Speech2Text` (genai API) |\n| LLM on device | ✘ | ✓ via `hailo_platform.genai.LLM` |\n| VLM on device | ✘ | ✓ via `Backend` (VLM chat) |\n| TTS (Piper) | ✓ CPU | ✓ CPU |\n| Full voice assistant | STT + CPU LLM + TTS | STT + on-device LLM + TTS |\n\n## Reference Implementations\n\nStudy these:\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fvoice_assistant\u002F` — Full voice + LLM assistant (Hailo-10H)\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fsimple_whisper_chat\u002F` — Simple STT example (Hailo-10H)\n- `hailo_apps\u002Fpython\u002Fstandalone_apps\u002Fspeech_recognition\u002F` — STT for **all Hailo devices** (8\u002F8L\u002F10H) using InferModel API:\n  - `speech_recognition.py` — Main app: mic recording, audio preprocessing, transcription loop\n  - `whisper_pipeline.py` — `WhisperPipeline` class: encoder+decoder inference via `InferModel`\n  - `audio_utils.py` — Audio recording, mel spectrogram, file I\u002FO\n  - `postprocessing.py` — Repetition penalty and token decoding\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fgen_ai_utils\u002Fvoice_processing\u002F` — Voice utilities (Hailo-10H):\n  - `speech_to_text.py` — `SpeechToTextProcessor` (Whisper via genai API)\n  - `text_to_speech.py` — `TextToSpeechProcessor` (Piper on CPU)\n  - `audio_recorder.py` — `AudioRecorder` (microphone capture)\n  - `vad.py` — Voice Activity Detection\n  - `interaction.py` — `VoiceInteractionManager` (high-level orchestrator)\n\n## Build Process\n\n### Step 1: Create App Directory\n\nCreate the app directory:\n\n```\nhailo_apps\u002Fpython\u002F\u003Ctype>\u002F\u003Capp_name>\u002F\n├── app.yaml              # App manifest (type: gen_ai)\n├── run.sh                # Launch wrapper\n├── __init__.py\n├── \u003Capp_name>.py         # Main app\n└── README.md             # Usage documentation (REQUIRED — never skip)\n```\n\nCreate `app.yaml` with `type: gen_ai` and `run.sh` wrapper.\nDo NOT register in `defines.py` or `resources_config.yaml`.\n\n### Step 2: Build Main App (Hailo-10H: Voice + LLM)\n\n```python\nimport signal\nimport threading\nfrom contextlib import redirect_stderr\nfrom io import StringIO\n\nfrom hailo_platform import VDevice\nfrom hailo_platform.genai import LLM\n\nfrom hailo_apps.python.core.common.hailo_logger import get_logger\nfrom hailo_apps.python.core.common.core import resolve_hef_path\nfrom hailo_apps.python.core.common.parser import get_standalone_parser\nfrom hailo_apps.python.core.common.defines import (\n    SHARED_VDEVICE_GROUP_ID,\n    HAILO10H_ARCH,\n)\n\nlogger = get_logger(__name__)\n\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.speech_to_text import SpeechToTextProcessor\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.text_to_speech import TextToSpeechProcessor\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.interaction import VoiceInteractionManager\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.vad import add_vad_args\n\nAPP_NAME = \"my_voice_app\"\n\nlogger = get_logger(__name__)\n\nAPP_NAME = MY_VOICE_APP\nSYSTEM_PROMPT = \"You are a helpful voice assistant. Keep responses concise and natural.\"\n\n\ndef main():\n    parser = get_standalone_parser()\n    parser.add_argument(\"--no-tts\", action=\"store_true\", help=\"Disable TTS (text only)\")\n    parser.add_argument(\"--system-prompt\", type=str, default=SYSTEM_PROMPT)\n    add_vad_args(parser)\n    args = parser.parse_args()\n\n    abort_event = threading.Event()\n    signal.signal(signal.SIGINT, lambda s, f: abort_event.set())\n\n    # VDevice\n    params = VDevice.create_params()\n    params.group_id = SHARED_VDEVICE_GROUP_ID\n    vdevice = VDevice(params)\n\n    # STT (Whisper on Hailo)\n    whisper_hef = resolve_hef_path(None, \"whisper\", arch=HAILO10H_ARCH)\n    with redirect_stderr(StringIO()):  # Suppress ALSA noise\n        stt = SpeechToTextProcessor(vdevice, str(whisper_hef))\n\n    # LLM (on Hailo)\n    llm_hef = resolve_hef_path(args.hef_path, APP_NAME, arch=HAILO10H_ARCH)\n    llm = LLM(vdevice, str(llm_hef))\n\n    # TTS (Piper on CPU)\n    tts = None if args.no_tts else TextToSpeechProcessor()\n\n    # Voice interaction manager\n    vim = VoiceInteractionManager(stt, tts, abort_event)\n\n    logger.info(\"Voice assistant ready. Speak into your microphone.\")\n    print(\"Voice assistant ready. Press Ctrl+C to quit.\\n\")\n\n    try:\n        while not abort_event.is_set():\n            # Listen for speech\n            user_text = vim.listen()\n            if not user_text or abort_event.is_set():\n                continue\n\n            logger.info(\"User said: %s\", user_text)\n            print(f\"You: {user_text}\")\n\n            # Generate response\n            prompt = [\n                {\"role\": \"system\", \"content\": [{\"type\": \"text\", \"text\": args.system_prompt}]},\n                {\"role\": \"user\", \"content\": [{\"type\": \"text\", \"text\": user_text}]},\n            ]\n            response = llm.generate_all(\n                prompt=prompt, temperature=0.1, max_generated_tokens=150\n            )\n            llm.clear_context()\n\n            print(f\"Assistant: {response}\\n\")\n\n            # Speak response\n            if tts and not abort_event.is_set():\n                vim.speak(response)\n    finally:\n        llm.release()\n        vdevice.release()\n        logger.info(\"Cleanup complete\")\n\n\nif __name__ == \"__main__\":\n    main()\n```\n\n### Step 3: Validate (Hailo-10H)\n\n```bash\npython3 .github\u002Fscripts\u002Fvalidate_app.py hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fmy_voice_app --smoke-test\n```\n\n### Step 2b: Build Main App (Hailo-8\u002F8L: Speech Recognition)\n\nFor Hailo-8\u002F8L, use the `InferModel` API with separate encoder + decoder HEFs:\n\n```python\nimport signal\nimport sys\nimport time\n\nimport numpy as np\n\nfrom hailo_apps.python.core.common.toolbox import resolve_arch\nfrom hailo_apps.python.core.common.hailo_logger import get_logger\nfrom hailo_apps.python.core.common.core import resolve_hef_paths\nfrom hailo_apps.python.core.common.parser import get_standalone_parser\nfrom hailo_apps.python.core.common.defines import (\n    WHISPER_H8_APP, RESOURCES_ROOT_PATH_DEFAULT, RESOURCES_NPY_DIR_NAME,\n)\n\nlogger = get_logger(__name__)\n\nAPP_NAME = \"my_speech_app\"\n\n\ndef main():\n    parser = get_standalone_parser()\n    parser.add_argument(\"--audio\", type=str, help=\"Audio file to transcribe (skip mic)\")\n    parser.add_argument(\"--variant\", default=\"base\", choices=[\"base\", \"tiny\", \"tiny.en\"],\n                        help=\"Whisper model variant\")\n    parser.add_argument(\"--duration\", type=int, default=10, help=\"Max recording seconds\")\n    args = parser.parse_args()\n\n    arch = resolve_arch(args.arch)  # Auto-detect or --arch flag\n\n    # Resolve encoder + decoder HEFs (plural — Whisper uses two separate models)\n    hef_paths = resolve_hef_paths(None, WHISPER_H8_APP, arch=arch)\n    encoder_path = str(hef_paths[\"encoder\"])\n    decoder_path = str(hef_paths[\"decoder\"])\n\n    # Decoder assets directory\n    npy_dir = str(Path(RESOURCES_ROOT_PATH_DEFAULT) \u002F RESOURCES_NPY_DIR_NAME)\n\n    # WhisperPipeline — uses InferModel API, works on ALL Hailo devices\n    from .whisper_pipeline import WhisperPipeline\n    pipeline = WhisperPipeline(\n        encoder_path=encoder_path,\n        decoder_path=decoder_path,\n        variant=args.variant,\n        npy_dir=npy_dir,\n        add_embed=(arch != \"hailo10h\"),  # H8\u002F8L need host-side Add operator\n    )\n\n    if args.audio:\n        # Single file transcription\n        mel = preprocess_audio_file(args.audio, pipeline.get_chunk_length())\n        pipeline.send_data(mel)\n        text = pipeline.get_transcription()\n        print(f\"Transcription: {text}\")\n    else:\n        # Interactive mic loop\n        logger.info(\"Ready. Press Enter to start recording, Enter to stop. 'q' to quit.\")\n        while True:\n            cmd = input(\"Press Enter to record (q to quit): \").strip()\n            if cmd.lower() == \"q\":\n                break\n            audio = record_from_mic(duration=args.duration)\n            mel = audio_to_mel(audio, pipeline.get_chunk_length())\n            pipeline.send_data(mel)\n            text = pipeline.get_transcription()\n            print(f\"Transcription: {text}\")\n\n    pipeline.stop()\n    logger.info(\"Done\")\n\n\nif __name__ == \"__main__\":\n    main()\n```\n\n**Key differences from H10 pattern**:\n- Uses `resolve_hef_paths()` (plural) — returns dict with `encoder` and `decoder` keys\n- Uses `WhisperPipeline` with `InferModel` API — NOT `SpeechToTextProcessor`\n- `add_embed` flag: `True` for H8\u002F8L (host-side Add operator), `False` for H10\n- No LLM\u002FVLM on device — inference is CPU or external service\n- Dependencies: `pip install -e \".[speech-rec]\"` (torch, transformers, sounddevice, scipy)\n\n### Step 3b: Validate (Hailo-8\u002F8L)\n\n```bash\npython3 .github\u002Fscripts\u002Fvalidate_app.py hailo_apps\u002Fpython\u002Fstandalone_apps\u002Fmy_speech_app --smoke-test\n```\n\n## Critical Conventions\n\n1. **STT on Hailo, TTS on CPU**: Never reverse this — Whisper needs the accelerator, Piper is CPU-only\n2. **Two STT APIs**: `SpeechToTextProcessor` (genai) for H10, `WhisperPipeline` (InferModel) for H8\u002F8L\u002F10H\n3. **Architecture detection**: Use `detect_hailo_arch()` or `--arch` flag — never hardcode\n4. **ALSA noise**: Wrap audio init with `redirect_stderr(StringIO())`\n5. **Abort event**: `threading.Event()` for interrupting generation and speech\n6. **Init order**: VDevice → STT → LLM (H10 only) → TTS → VoiceInteractionManager\n7. **Cleanup order**: Reverse of init, in `finally` block\n8. **VAD args**: Always use `add_vad_args(parser)` for `--vad`, `--vad-aggressiveness`, `--vad-energy-threshold`\n9. **--no-tts**: Always support text-only mode as fallback\n10. **H8\u002F8L HEF resolution**: Use `resolve_hef_paths()` (plural) — Whisper needs encoder + decoder HEFs\n11. **H10 HEF resolution**: Use `resolve_hef_path()` (singular) — single HEF per model\n12. **Dependencies**: H10 voice requires `[gen-ai]` extras; H8\u002F8L STT requires `[speech-rec]` extras\n\n## Adding Voice to an Existing App\n\nTo add voice to any existing LLM\u002FVLM app:\n\n1. Import voice utilities\n2. Add `--voice` and `--no-tts` CLI flags + `add_vad_args(parser)`\n3. Init STT + TTS alongside existing model\n4. Replace `input()` with `vim.listen()` when `--voice` is set\n5. Add `vim.speak(response)` after generating output\n6. Add `abort_event` for interruption support\n\n","# Skill: Build Voice Application\n\nBuild voice apps for **all Hailo accelerators**: Whisper STT on Hailo-8\u002F8L\u002F10H + optional Piper TTS on CPU.\n\n## When This Skill Is Loaded\n\n- User wants **speech input** or **speech output** in a Hailo app\n- User mentions: voice, speech, Whisper, TTS, microphone, STT, speak, listen\n- User wants to add voice to an existing LLM or VLM app\n- User wants speech recognition on Hailo-8 or Hailo-8L\n\n## Hardware Compatibility\n\n| Feature | Hailo-8\u002F8L | Hailo-10H |\n|---|---|---|\n| STT (Whisper) | ✓ via `InferModel` (encoder+decoder HEFs) | ✓ via `Speech2Text` (genai API) |\n| LLM on device | ✘ | ✓ via `hailo_platform.genai.LLM` |\n| VLM on device | ✘ | ✓ via `Backend` (VLM chat) |\n| TTS (Piper) | ✓ CPU | ✓ CPU |\n| Full voice assistant | STT + CPU LLM + TTS | STT + on-device LLM + TTS |\n\n## Reference Implementations\n\nStudy these:\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fvoice_assistant\u002F` — Full voice + LLM assistant (Hailo-10H)\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fsimple_whisper_chat\u002F` — Simple STT example (Hailo-10H)\n- `hailo_apps\u002Fpython\u002Fstandalone_apps\u002Fspeech_recognition\u002F` — STT for **all Hailo devices** (8\u002F8L\u002F10H) using InferModel API:\n  - `speech_recognition.py` — Main app: mic recording, audio preprocessing, transcription loop\n  - `whisper_pipeline.py` — `WhisperPipeline` class: encoder+decoder inference via `InferModel`\n  - `audio_utils.py` — Audio recording, mel spectrogram, file I\u002FO\n  - `postprocessing.py` — Repetition penalty and token decoding\n- `hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fgen_ai_utils\u002Fvoice_processing\u002F` — Voice utilities (Hailo-10H):\n  - `speech_to_text.py` — `SpeechToTextProcessor` (Whisper via genai API)\n  - `text_to_speech.py` — `TextToSpeechProcessor` (Piper on CPU)\n  - `audio_recorder.py` — `AudioRecorder` (microphone capture)\n  - `vad.py` — Voice Activity Detection\n  - `interaction.py` — `VoiceInteractionManager` (high-level orchestrator)\n\n## Build Process\n\n### Step 1: Create App Directory\n\nCreate the app directory:\n\n```\nhailo_apps\u002Fpython\u002F\u003Ctype>\u002F\u003Capp_name>\u002F\n├── app.yaml              # App manifest (type: gen_ai)\n├── run.sh                # Launch wrapper\n├── __init__.py\n├── \u003Capp_name>.py         # Main app\n└── README.md             # Usage documentation (REQUIRED — never skip)\n```\n\nCreate `app.yaml` with `type: gen_ai` and `run.sh` wrapper.\nDo NOT register in `defines.py` or `resources_config.yaml`.\n\n### Step 2: Build Main App (Hailo-10H: Voice + LLM)\n\n```python\nimport signal\nimport threading\nfrom contextlib import redirect_stderr\nfrom io import StringIO\n\nfrom hailo_platform import VDevice\nfrom hailo_platform.genai import LLM\n\nfrom hailo_apps.python.core.common.hailo_logger import get_logger\nfrom hailo_apps.python.core.common.core import resolve_hef_path\nfrom hailo_apps.python.core.common.parser import get_standalone_parser\nfrom hailo_apps.python.core.common.defines import (\n    SHARED_VDEVICE_GROUP_ID,\n    HAILO10H_ARCH,\n)\n\nlogger = get_logger(__name__)\n\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.speech_to_text import SpeechToTextProcessor\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.text_to_speech import TextToSpeechProcessor\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.interaction import VoiceInteractionManager\nfrom hailo_apps.python.gen_ai_apps.gen_ai_utils.voice_processing.vad import add_vad_args\n\nAPP_NAME = \"my_voice_app\"\n\nlogger = get_logger(__name__)\n\nAPP_NAME = MY_VOICE_APP\nSYSTEM_PROMPT = \"You are a helpful voice assistant. Keep responses concise and natural.\"\n\n\ndef main():\n    parser = get_standalone_parser()\n    parser.add_argument(\"--no-tts\", action=\"store_true\", help=\"Disable TTS (text only)\")\n    parser.add_argument(\"--system-prompt\", type=str, default=SYSTEM_PROMPT)\n    add_vad_args(parser)\n    args = parser.parse_args()\n\n    abort_event = threading.Event()\n    signal.signal(signal.SIGINT, lambda s, f: abort_event.set())\n\n    # VDevice\n    params = VDevice.create_params()\n    params.group_id = SHARED_VDEVICE_GROUP_ID\n    vdevice = VDevice(params)\n\n    # STT (Whisper on Hailo)\n    whisper_hef = resolve_hef_path(None, \"whisper\", arch=HAILO10H_ARCH)\n    with redirect_stderr(StringIO()):  # Suppress ALSA noise\n        stt = SpeechToTextProcessor(vdevice, str(whisper_hef))\n\n    # LLM (on Hailo)\n    llm_hef = resolve_hef_path(args.hef_path, APP_NAME, arch=HAILO10H_ARCH)\n    llm = LLM(vdevice, str(llm_hef))\n\n    # TTS (Piper on CPU)\n    tts = None if args.no_tts else TextToSpeechProcessor()\n\n    # Voice interaction manager\n    vim = VoiceInteractionManager(stt, tts, abort_event)\n\n    logger.info(\"Voice assistant ready. Speak into your microphone.\")\n    print(\"Voice assistant ready. Press Ctrl+C to quit.\\n\")\n\n    try:\n        while not abort_event.is_set():\n            # Listen for speech\n            user_text = vim.listen()\n            if not user_text or abort_event.is_set():\n                continue\n\n            logger.info(\"User said: %s\", user_text)\n            print(f\"You: {user_text}\")\n\n            # Generate response\n            prompt = [\n                {\"role\": \"system\", \"content\": [{\"type\": \"text\", \"text\": args.system_prompt}]},\n                {\"role\": \"user\", \"content\": [{\"type\": \"text\", \"text\": user_text}]},\n            ]\n            response = llm.generate_all(\n                prompt=prompt, temperature=0.1, max_generated_tokens=150\n            )\n            llm.clear_context()\n\n            print(f\"Assistant: {response}\\n\")\n\n            # Speak response\n            if tts and not abort_event.is_set():\n                vim.speak(response)\n    finally:\n        llm.release()\n        vdevice.release()\n        logger.info(\"Cleanup complete\")\n\n\nif __name__ == \"__main__\":\n    main()\n```\n\n### Step 3: Validate (Hailo-10H)\n\n```bash\npython3 .github\u002Fscripts\u002Fvalidate_app.py hailo_apps\u002Fpython\u002Fgen_ai_apps\u002Fmy_voice_app --smoke-test\n```\n\n### Step 2b: Build Main App (Hailo-8\u002F8L: Speech Recognition)\n\nFor Hailo-8\u002F8L, use the `InferModel` API with separate encoder + decoder HEFs:\n\n```python\nimport signal\nimport sys\nimport time\n\nimport numpy as np\n\nfrom hailo_apps.python.core.common.toolbox import resolve_arch\nfrom hailo_apps.python.core.common.hailo_logger import get_logger\nfrom hailo_apps.python.core.common.core import resolve_hef_paths\nfrom hailo_apps.python.core.common.parser import get_standalone_parser\nfrom hailo_apps.python.core.common.defines import (\n    WHISPER_H8_APP, RESOURCES_ROOT_PATH_DEFAULT, RESOURCES_NPY_DIR_NAME,\n)\n\nlogger = get_logger(__name__)\n\nAPP_NAME = \"my_speech_app\"\n\n\ndef main():\n    parser = get_standalone_parser()\n    parser.add_argument(\"--audio\", type=str, help=\"Audio file to transcribe (skip mic)\")\n    parser.add_argument(\"--variant\", default=\"base\", choices=[\"base\", \"tiny\", \"tiny.en\"],\n                        help=\"Whisper model variant\")\n    parser.add_argument(\"--duration\", type=int, default=10, help=\"Max recording seconds\")\n    args = parser.parse_args()\n\n    arch = resolve_arch(args.arch)  # Auto-detect or --arch flag\n\n    # Resolve encoder + decoder HEFs (plural — Whisper uses two separate models)\n    hef_paths = resolve_hef_paths(None, WHISPER_H8_APP, arch=arch)\n    encoder_path = str(hef_paths[\"encoder\"])\n    decoder_path = str(hef_paths[\"decoder\"])\n\n    # Decoder assets directory\n    npy_dir = str(Path(RESOURCES_ROOT_PATH_DEFAULT) \u002F RESOURCES_NPY_DIR_NAME)\n\n    # WhisperPipeline — uses InferModel API, works on ALL Hailo devices\n    from .whisper_pipeline import WhisperPipeline\n    pipeline = WhisperPipeline(\n        encoder_path=encoder_path,\n        decoder_path=decoder_path,\n        variant=args.variant,\n        npy_dir=npy_dir,\n        add_embed=(arch != \"hailo10h\"),  # H8\u002F8L need host-side Add operator\n    )\n\n    if args.audio:\n        # Single file transcription\n        mel = preprocess_audio_file(args.audio, pipeline.get_chunk_length())\n        pipeline.send_data(mel)\n        text = pipeline.get_transcription()\n        print(f\"Transcription: {text}\")\n    else:\n        # Interactive mic loop\n        logger.info(\"Ready. Press Enter to start recording, Enter to stop. 'q' to quit.\")\n        while True:\n            cmd = input(\"Press Enter to record (q to quit): \").strip()\n            if cmd.lower() == \"q\":\n                break\n            audio = record_from_mic(duration=args.duration)\n            mel = audio_to_mel(audio, pipeline.get_chunk_length())\n            pipeline.send_data(mel)\n            text = pipeline.get_transcription()\n            print(f\"Transcription: {text}\")\n\n    pipeline.stop()\n    logger.info(\"Done\")\n\n\nif __name__ == \"__main__\":\n    main()\n```\n\n**Key differences from H10 pattern**:\n- Uses `resolve_hef_paths()` (plural) — returns dict with `encoder` and `decoder` keys\n- Uses `WhisperPipeline` with `InferModel` API — NOT `SpeechToTextProcessor`\n- `add_embed` flag: `True` for H8\u002F8L (host-side Add operator), `False` for H10\n- No LLM\u002FVLM on device — inference is CPU or external service\n- Dependencies: `pip install -e \".[speech-rec]\"` (torch, transformers, sounddevice, scipy)\n\n### Step 3b: Validate (Hailo-8\u002F8L)\n\n```bash\npython3 .github\u002Fscripts\u002Fvalidate_app.py hailo_apps\u002Fpython\u002Fstandalone_apps\u002Fmy_speech_app --smoke-test\n```\n\n## Critical Conventions\n\n1. **STT on Hailo, TTS on CPU**: Never reverse this — Whisper needs the accelerator, Piper is CPU-only\n2. **Two STT APIs**: `SpeechToTextProcessor` (genai) for H10, `WhisperPipeline` (InferModel) for H8\u002F8L\u002F10H\n3. **Architecture detection**: Use `detect_hailo_arch()` or `--arch` flag — never hardcode\n4. **ALSA noise**: Wrap audio init with `redirect_stderr(StringIO())`\n5. **Abort event**: `threading.Event()` for interrupting generation and speech\n6. **Init order**: VDevice → STT → LLM (H10 only) → TTS → VoiceInteractionManager\n7. **Cleanup order**: Reverse of init, in `finally` block\n8. **VAD args**: Always use `add_vad_args(parser)` for `--vad`, `--vad-aggressiveness`, `--vad-energy-threshold`\n9. **--no-tts**: Always support text-only mode as fallback\n10. **H8\u002F8L HEF resolution**: Use `resolve_hef_paths()` (plural) — Whisper needs encoder + decoder HEFs\n11. **H10 HEF resolution**: Use `resolve_hef_path()` (singular) — single HEF per model\n12. **Dependencies**: H10 voice requires `[gen-ai]` extras; H8\u002F8L STT requires `[speech-rec]` extras\n\n## Adding Voice to an Existing App\n\nTo add voice to any existing LLM\u002FVLM app:\n\n1. Import voice utilities\n2. Add `--voice` and `--no-tts` CLI flags + `add_vad_args(parser)`\n3. Init STT + TTS alongside existing model\n4. Replace `input()` with `vim.listen()` when `--voice` is set\n5. Add `vim.speak(response)` after generating output\n6. Add `abort_event` for interruption support\n\n",{},"main","b285906ff105666c6100208690d89f76f2c51e45",".github\u002Fskills\u002Fhl-build-voice-app",0,"https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat","https:\u002F\u002Fgithub.com\u002FDatnewlevel","Embedded Developer | datnewlevel's workspace","2026-07-21 23:34:00",{"source":7,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add Datnewlevel\u002FA.I-drone-pi5-hailo-hat --skill hl-build-voice-app",{"tool":36,"command":37},"bunx","bunx skills add Datnewlevel\u002FA.I-drone-pi5-hailo-hat --skill hl-build-voice-app",{"tool":39,"command":40},"pnpm","pnpm dlx skills add Datnewlevel\u002FA.I-drone-pi5-hailo-hat --skill hl-build-voice-app","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-voice-app\u002FSKILL.md",true,[],[45,52,59,66,73,80],{"skillNo":46,"name":47,"nameZh":6,"description":48,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":49,"categories":50,"keywords":51,"publishedAt":17},"01M79H0V6G7WR1THHEJYK0E9Q5","hl-build-vlm-app","Build a complete Vision-Language Model application that uses the Hailo-10H VLM for image understanding.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-vlm-app\u002FSKILL.md",[],[],{"skillNo":53,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":17},"01M79H0S34YW4NPQQNM4TCQ48D","hl-build-standalone-app","Build a complete standalone app using HailoInfer + OpenCV on Hailo-8\u002F8L\u002F10H.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-standalone-app\u002FSKILL.md",[],[],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":17},"01M79H0NH2XZZZ3XME7KGM9Z09","hl-build-pipeline-app","Build a complete GStreamer pipeline app for real-time video processing on Hailo-8\u002F8L\u002F10H.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-pipeline-app\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":17},"01M79H0KGH250T5B4HQYNP88FD","hl-build-llm-app","Build a complete LLM text generation app running on Hailo-10H.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-llm-app\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":17},"01M79H0DP7258FPC3FR36SWMF0","hl-build-agent-app","Build a complete agent app with LLM reasoning + tool execution on Hailo-10H.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.github\u002Fskills\u002Fhl-build-agent-app\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":17},"01M79H0CP44JQRYMQEYKC6M8YB","hl-validate","Validation at every phase gate.","https:\u002F\u002Fgithub.com\u002FDatnewlevel\u002FA.I-drone-pi5-hailo-hat\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fhl-validate\u002FSKILL.md",[],[]]