modify project structure and add download_models.py

2024-05-01 13:35:00 +07:00
parent 8504126d4a
commit a2375e6daf
19 changed files with 36 additions and 9 deletions
@@ -0,0 +1,85 @@
+import json
+import os
+import struct
+import time
+
+import noisereduce as nr
+import pvporcupine
+import vosk
+import yaml
+from fuzzywuzzy import fuzz
+from pvrecorder import PvRecorder
+
+from data import config
+from utils import download_models, execute_cmd, play
+
+
+class Jarvis:
+    def __init__(self):
+        download_models.install_vosk_model()
+        self.recorder = None
+        self.CDIR = os.getcwd()
+        self.VA_CMD_LIST = yaml.safe_load(open('data/commands.yaml', encoding='utf8'))
+        self.porcupine = pvporcupine.create(
+            access_key=config.PICOVOICE_TOKEN,
+            keywords=['jarvis'],
+            sensitivities=[1]
+        )
+        self.kaldi_rec = vosk.KaldiRecognizer(vosk.Model("data/model_small"), 16000)
+
+    def main(self):
+        self.recorder = PvRecorder(
+            device_index=config.MICROPHONE_INDEX,
+            frame_length=self.porcupine.frame_length
+        )
+        self.recorder.start()
+        self.play("run")
+        time.sleep(0.5)
+        ltc = time.time() - 1000
+        while True:
+            try:
+                pcm = self.recorder.read()
+                reduced_audio = nr.reduce_noise(
+                    y=pcm,
+                    sr=16000,
+                    prop_decrease=0.6
+                )
+                if self.porcupine.process(reduced_audio) >= 0:
+                    self.recorder.stop()
+                    self.play("greet", True)
+                    self.recorder.start()
+                    ltc = time.time()
+                while time.time() - ltc <= 10:
+                    pcm = self.recorder.read()
+                    sp = struct.pack("h" * len(pcm), *pcm)
+                    if self.kaldi_rec.AcceptWaveform(sp):
+                        if self.va_respond(json.loads(self.kaldi_rec.Result())["text"]):
+                            ltc = time.time()
+                        break
+            except Exception as err:
+                print(f"Unexpected {err=}, {type(err)=}")
+                raise
+
+    def va_respond(self, voice: str):
+        print(f"Распознано: {voice}")
+        for x in config.VA_ALIAS + config.VA_TBR:
+            voice = voice.replace(x, "").strip()
+        rc = {'cmd': '', 'percent': 0}
+        for c, v in self.VA_CMD_LIST.items():
+            for x in v:
+                vrt = fuzz.partial_ratio(voice.lower(), x.lower())
+                if vrt > rc['percent']:
+                    rc['cmd'] = c
+                    rc['percent'] = vrt
+        if len(rc['cmd'].strip()) <= 0:
+            return False
+        elif rc['percent'] < 70 or rc['cmd'] not in self.VA_CMD_LIST.keys():
+            self.play("not_found")
+            time.sleep(1)
+            return False
+        else:
+            execute_cmd.execute_cmd(self, rc['cmd'])
+            return True
+
+    def play(self, phrase, wait_done=True):
+        play.play(self, phrase, wait_done)