Refactor tool-speechtotext: extract sttlib shared library and add tests

Extract duplicated code (Whisper loading, audio recording, transcription,
VAD processing) into reusable sttlib/ package. Rewrite all 3 scripts as
thin wrappers. Add 24 unit tests with mocked hardware. Fix GPU fallback
bug in assistant.py and args.system assignment bug.

python/tool-speechtotext/voice_to_terminal.py

@@ -1,31 +1,16 @@
-import sounddevice as sd
-import numpy as np
-import pyperclip
-import sys
 import argparse
-import os
 import subprocess
-import ollama
 import json
-from faster_whisper import WhisperModel
+import ollama
+from sttlib import load_whisper_model, record_until_enter, transcribe
 
 # --- Configuration ---
-os.environ["CT2_CUDA_ALLOW_FP16"] = "1"
-MODEL_SIZE = "medium"
 OLLAMA_MODEL = "qwen2.5-coder:7b"
 CONFIRM_COMMANDS = True  # Set to False to run commands instantly
 
-# Load Whisper on GPU
-print("Loading Whisper model...")
-try:
-    model = WhisperModel(MODEL_SIZE, device="cuda", compute_type="float16")
-except Exception as e:
-    print(f"Error loading GPU: {e}, falling back to CPU")
-    model = WhisperModel(MODEL_SIZE, device="cpu", compute_type="int8")
 
 # --- Terminal Tool ---
 
-
 def run_terminal_command(command: str):
     """
     Executes a bash command in the Linux terminal.
@@ -33,8 +18,7 @@ def run_terminal_command(command: str):
     """
     if CONFIRM_COMMANDS:
         print(f"\n{'='*40}")
-        print(f"⚠️  AI SUGGESTED: \033[1;32m{command}\033[0m")
-        # Allow user to provide feedback if they say 'n'
+        print(f"\u26a0\ufe0f  AI SUGGESTED: \033[1;32m{command}\033[0m")
         choice = input("   Confirm? [Y/n] or provide feedback: ").strip()
 
         if choice.lower() == 'n':
@@ -57,22 +41,15 @@ def run_terminal_command(command: str):
         return f"Execution Error: {str(e)}"
 
 
-def record_audio():
-    fs, recording = 16000, []
-    print("\n[READY] Press Enter to START...")
-    input()
-    print("[RECORDING] Press Enter to STOP...")
-    def cb(indata, f, t, s): recording.append(indata.copy())
-    with sd.InputStream(samplerate=fs, channels=1, callback=cb):
-        input()
-    return np.concatenate(recording, axis=0)
-
-
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--model", default=OLLAMA_MODEL)
+    parser.add_argument("--model-size", default="medium",
+                        help="Whisper model size")
     args, _ = parser.parse_known_args()
 
+    whisper_model = load_whisper_model(args.model_size)
+
     # Initial System Prompt
     messages = [{
         'role': 'system',
@@ -88,9 +65,8 @@ def main():
     while True:
         try:
             # 1. Voice Capture
-            audio_data = record_audio()
-            segments, _ = model.transcribe(audio_data.flatten(), beam_size=5)
-            user_text = "".join([s.text for s in segments]).strip()
+            audio_data = record_until_enter()
+            user_text = transcribe(whisper_model, audio_data.flatten())
             if not user_text:
                 continue