Drafting with Abominable Intelligence

.gitignore

@@ -1,3 +1,4 @@
 .env
 node_modules/
-.venv/
+.venv/
+src/scripts/py/vosk-model*/

package-lock.json

@@ -9,6 +9,7 @@
       "version": "1.0.0",
       "license": "ISC",
       "dependencies": {
+        "mic": "^2.1.2",
         "redis": "^6.1.0"
       },
       "devDependencies": {
@@ -320,6 +321,12 @@
         "node": ">=0.12.0"
       }
     },
+    "node_modules/mic": {
+      "version": "2.1.2",
+      "resolved": "https://registry.npmjs.org/mic/-/mic-2.1.2.tgz",
+      "integrity": "sha512-rpl4tgdXX24sAzYwjRc5OZfGNAuhUIIjdd0cw8+Ubq7rp3iGhi40AdqcwurDWhEZADk60tPOxb3E2MpoeLeyxw==",
+      "license": "MIT"
+    },
     "node_modules/minimatch": {
       "version": "10.2.5",
       "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-10.2.5.tgz",

package.json

@@ -12,6 +12,7 @@
   "license": "ISC",
   "type": "commonjs",
   "dependencies": {
+    "mic": "^2.1.2",
     "redis": "^6.1.0"
   },
   "devDependencies": {

src/scripts/py/config.json

@@ -0,0 +1,11 @@
+{
+  "wakeword": "alexa",
+  "wakewordThreshold": 0.80,
+  "cooldownSeconds": 2.5,
+  "confirmFrames": 3,
+  "listenSeconds": 12,
+  "silenceThreshold": 0.015,
+  "silenceFrames": 10,
+  "voskModelPath": "./vosk-model-small-en-us-0.15",
+  "sampleRate": 16000
+}

src/scripts/py/info.txt

@@ -0,0 +1 @@
+PyAudio requires Python version <=3.13

src/scripts/py/requirements.txt

@@ -0,0 +1,4 @@
+openwakeword
+onnxruntime
+numpy
+pyaudio

src/scripts/py/transcribe.js

@@ -0,0 +1,46 @@
+const { spawn } = require('child_process');
+const fs = require('fs');
+
+const config = JSON.parse(fs.readFileSync('./config.json', 'utf8'));
+
+const py = spawn('python', ['wakeword.py', JSON.stringify(config)], {
+  stdio: ['ignore', 'pipe', 'pipe']
+});
+
+let buffer = '';
+
+py.stdout.on('data', (data) => {
+  buffer += data.toString();
+  const lines = buffer.split('\n');
+  buffer = lines.pop();
+
+  for (const line of lines) {
+    if (!line.trim()) continue;
+
+    try {
+      const msg = JSON.parse(line);
+
+      if (msg.event === 'ready') {
+        console.log('Python ready');
+      } else if (msg.event === 'wakeword') {
+        console.log(`Wakeword detected: ${msg.model} (${msg.score})`);
+      } else if (msg.event === 'partial') {
+        console.log('Partial:', msg.text);
+      } else if (msg.event === 'text' || msg.event === 'final') {
+        console.log('Transcript:', msg.text);
+      } else if (msg.event === 'done') {
+        console.log('Back to wakeword mode');
+      }
+    } catch {
+      console.log('[PY]', line);
+    }
+  }
+});
+
+py.stderr.on('data', (data) => {
+  console.error('[PY ERR]', data.toString());
+});
+
+py.on('close', (code) => {
+  console.log('Python exited with code', code);
+});

src/scripts/py/wakeword.py

@@ -0,0 +1,112 @@
+import json
+import sys
+import time
+import numpy as np
+import pyaudio
+import vosk
+from openwakeword.model import Model
+
+CONFIG = json.loads(sys.argv[1])
+
+RATE = CONFIG.get("sampleRate", 16000)
+CHUNK = 1280
+WAKEWORD = CONFIG.get("wakeword", "alexa")
+THRESHOLD = CONFIG.get("wakewordThreshold", 0.75)
+COOLDOWN = CONFIG.get("cooldownSeconds", 2.5)
+CONFIRM_FRAMES = CONFIG.get("confirmFrames", 3)
+LISTEN_SECONDS = CONFIG.get("listenSeconds", 12)
+SILENCE_THRESHOLD = CONFIG.get("silenceThreshold", 0.02)
+SILENCE_FRAMES = CONFIG.get("silenceFrames", 12)
+VOSK_MODEL_PATH = CONFIG["voskModelPath"]
+
+vosk.SetLogLevel(-1)
+vosk_model = vosk.Model(VOSK_MODEL_PATH)
+recognizer = vosk.KaldiRecognizer(vosk_model, RATE)
+recognizer.SetWords(True)
+
+oww = Model(wakeword_models=[WAKEWORD], inference_framework="onnx")
+
+audio = pyaudio.PyAudio()
+stream = audio.open(
+    format=pyaudio.paInt16,
+    channels=1,
+    rate=RATE,
+    input=True,
+    frames_per_buffer=CHUNK,
+)
+
+print(json.dumps({"event": "ready"}), flush=True)
+
+mode = "wakeword"
+high_count = 0
+last_trigger = 0
+silence_count = 0
+listen_deadline = None
+
+def rms_level(samples):
+    x = samples.astype(np.float32) / 32768.0
+    return np.sqrt(np.mean(x * x))
+
+try:
+    while True:
+        data = stream.read(CHUNK, exception_on_overflow=False)
+        samples = np.frombuffer(data, dtype=np.int16)
+        now = time.time()
+
+        if mode == "wakeword":
+            preds = oww.predict(samples)
+            score = preds.get(WAKEWORD, 0.0)
+
+            if score >= THRESHOLD:
+                high_count += 1
+            else:
+                high_count = 0
+
+            if high_count >= CONFIRM_FRAMES and (now - last_trigger) > COOLDOWN:
+                last_trigger = now
+                high_count = 0
+                print(json.dumps({
+                    "event": "wakeword",
+                    "model": WAKEWORD,
+                    "score": float(score)
+                }), flush=True)
+                mode = "transcribe"
+                recognizer.Reset()
+                silence_count = 0
+                listen_deadline = now + LISTEN_SECONDS
+
+        else:
+            if recognizer.AcceptWaveform(data):
+                result = json.loads(recognizer.Result())
+                if result.get("text"):
+                    print(json.dumps({"event": "final", "text": result["text"]}), flush=True)
+                    listen_deadline = now + LISTEN_SECONDS
+                    silence_count = 0
+            else:
+                partial = json.loads(recognizer.PartialResult()).get("partial", "")
+                if partial:
+                    print(json.dumps({"event": "partial", "text": partial}), flush=True)
+
+            level = rms_level(samples)
+            if level < SILENCE_THRESHOLD:
+                silence_count += 1
+            else:
+                silence_count = 0
+
+            if silence_count >= SILENCE_FRAMES or now > listen_deadline:
+                final = json.loads(recognizer.FinalResult())
+                if final.get("text"):
+                    print(json.dumps({"event": "final", "text": final["text"]}), flush=True)
+
+                print(json.dumps({"event": "done"}), flush=True)
+                mode = "wakeword"
+                silence_count = 0
+                high_count = 0
+                listen_deadline = None
+
+except KeyboardInterrupt:
+    pass
+finally:
+    stream.stop_stream()
+    stream.close()
+    audio.terminate()