From a4b919c11eea3316ef2e15d5a09649831b6156a6 Mon Sep 17 00:00:00 2001
From: Bossiara13 <236771060+DmitryBykov-ISPO@users.noreply.github.com>
Date: Wed, 22 Apr 2026 17:46:44 +0300
Subject: [PATCH] refactor: replace porcupine wake-word with vosk
 grammar-constrained recognizer

uses a second KaldiRecognizer with a JSON grammar of WAKE_WORDS + [unk] for
the wake phase, then hands off to the full-vocab recognizer for the 10s
command window. drops the pvporcupine dependency entirely.
---
 config.py | 10 ++++++----
 main.py   | 52 +++++++++++++++++++++++++++++-----------------------
 2 files changed, 35 insertions(+), 27 deletions(-)

diff --git a/config.py b/config.py
index 28a9710..01f9f2c 100644
--- a/config.py
+++ b/config.py
@@ -6,10 +6,15 @@ load_dotenv("dev.env")
 
 # Конфигурация
 VA_NAME = 'Jarvis'
-VA_VER = "0.1.0"
+VA_VER = "0.2.0"
 VA_ALIAS = ('джарвис',)
 VA_TBR = ('скажи', 'покажи', 'ответь', 'произнеси', 'расскажи', 'сколько', 'слушай')
 
+# Wake-words распознаются Vosk'ом с grammar-constraint. Добавляйте сюда
+# варианты написания — Vosk с русской моделью может расшифровать "jarvis"
+# как "джарвис", "жарвис" и т.п.
+WAKE_WORDS = ('jarvis', 'джарвис')
+
 # ID микрофона (можете просто менять ID пока при запуске не отобразится нужный)
 # -1 это стандартное записывающее устройство
 MICROPHONE_INDEX = -1
@@ -17,9 +22,6 @@ MICROPHONE_INDEX = -1
 # Путь к браузеру Google Chrome
 CHROME_PATH = 'C:/Program Files (x86)/Google/Chrome/Application/chrome.exe %s'
 
-# Токен Picovoice
-PICOVOICE_TOKEN = os.getenv('PICOVOICE_TOKEN')
-
 # Токен Groq
 GROQ_TOKEN = os.getenv('GROQ_TOKEN')
 GROQ_BASE_URL = "https://api.groq.com/openai/v1"
diff --git a/main.py b/main.py
index dc7208c..52aa60a 100644
--- a/main.py
+++ b/main.py
@@ -11,7 +11,6 @@ from ctypes import POINTER, cast
 
 import openai
 from openai import OpenAI
-import pvporcupine
 import simpleaudio as sa
 import vosk
 import yaml
@@ -39,18 +38,14 @@ message_log = [system_message]
 
 client = OpenAI(api_key=config.GROQ_TOKEN, base_url=config.GROQ_BASE_URL)
 
-# PORCUPINE
-porcupine = pvporcupine.create(
-    access_key=config.PICOVOICE_TOKEN,
-    keywords=['jarvis'],
-    sensitivities=[1]
-)
-# print(pvporcupine.KEYWORDS)
-
-# VOSK
 model = vosk.Model("model_small")
 samplerate = 16000
 device = config.MICROPHONE_INDEX
+
+# wake-word phase uses a grammar-constrained recognizer so only WAKE_WORDS
+# (plus [unk] filler) can match — command phase needs full vocab, hence two.
+wake_grammar = json.dumps(list(config.WAKE_WORDS) + ["[unk]"], ensure_ascii=False)
+wake_rec = vosk.KaldiRecognizer(model, samplerate, wake_grammar)
 kaldi_rec = vosk.KaldiRecognizer(model, samplerate)
 q = queue.Queue()
 
@@ -268,33 +263,42 @@ def execute_cmd(cmd: str, voice: str):
 
     elif cmd == 'off':
         play("off", True)
-
-        porcupine.delete()
         exit(0)
 
 
-# `-1` is the default input audio device.
-recorder = PvRecorder(device_index=config.MICROPHONE_INDEX, frame_length=porcupine.frame_length)
+recorder = PvRecorder(device_index=config.MICROPHONE_INDEX, frame_length=512)
 recorder.start()
 print('Using device: %s' % recorder.selected_device)
 
-print(f"Jarvis (v0.1.0) начал свою работу ...")
+print(f"Jarvis (v{config.VA_VER}) начал свою работу ...")
 play("run")
 time.sleep(0.5)
 
-ltc = time.time() - 1000
+
+def heard_wake_word(text: str) -> bool:
+    if not text:
+        return False
+    lowered = text.lower()
+    return any(w in lowered for w in config.WAKE_WORDS)
+
 
 while True:
     try:
         pcm = recorder.read()
-        keyword_index = porcupine.process(pcm)
+        sp = struct.pack("h" * len(pcm), *pcm)
 
-        if keyword_index >= 0:
-            recorder.stop()
-            play("greet", True)
-            print("Yes, sir.")
-            recorder.start()  # prevent self recording
-            ltc = time.time()
+        if not wake_rec.AcceptWaveform(sp):
+            continue
+
+        result_text = json.loads(wake_rec.Result()).get("text", "")
+        if not heard_wake_word(result_text):
+            continue
+
+        play("greet", True)
+        print("Yes, sir.")
+        # reset the command recognizer so stale audio doesn't bleed into it
+        kaldi_rec.Reset()
+        ltc = time.time()
 
         while time.time() - ltc <= 10:
             pcm = recorder.read()
@@ -306,6 +310,8 @@ while True:
 
                 break
 
+        wake_rec.Reset()
+
     except Exception as err:
         print(f"Unexpected {err=}, {type(err)=}")
         raise