New image generation model and LLMs

V1.7.1/README.md

@@ -0,0 +1,82 @@
+# V1.7 Speech Agent
+
+A voice-driven AI assistant that listens for spoken queries, performs local actions, optionally captures images for visual requests, and answers using Ollama chat models.
+
+## Features
+
+- Voice input with Whisper speech recognition
+- Text-to-speech output using `edge-tts`
+- Intelligent command handling with Ollama chat models
+- App launching on macOS for `open <app>` voice commands
+- Web search query simplification and results retrieval
+- Vision-context detection and camera image capture for visual queries
+- ANSI-enhanced console output with `rich`
+
+## Requirements
+
+- macOS
+- Python 3.11+ (recommended)
+- Microphone access
+- Camera access for image-based queries
+- Installed Python packages:
+  - `ollama`
+  - `speech_recognition`
+  - `edge_tts`
+  - `ddgs`
+  - `opencv-python`
+  - `rich`
+
+## Installation
+
+1. Clone or copy this folder:
+
+```bash
+cd /path/to/AI/OSS_speech_model/Phase-1/V1.7
+```
+
+2. Create and activate a Python virtual environment:
+
+```bash
+python3 -m venv venv
+source venv/bin/activate
+```
+
+3. Install required packages:
+
+```bash
+pip install ollama speech_recognition edge_tts ddgs opencv-python rich
+```
+
+4. Make sure your environment has access to Ollama.
+
+## Usage
+
+Run the assistant with:
+
+```bash
+python main.py
+```
+
+The assistant will prompt:
+
+- Choose a voice: `Male` or `Female`
+- Speak a query after the prompt
+- It will attempt to answer, launch apps, or capture an image if visual context is needed
+- When finished, it asks whether you want to continue
+
+## Notes
+
+- The current implementation uses `afplay` to play audio on macOS.
+- App launching relies on `mdfind` and the local macOS application metadata.
+- If the assistant cannot understand audio, it prompts again.
+- Visual queries are captured and saved temporarily as `instant_photo.png` in the current folder.
+
+## Troubleshooting
+
+- If speech recognition fails, verify microphone permissions and `speech_recognition` installation.
+- If camera capture fails, verify camera permissions and that `opencv-python` can access the device.
+- If audio playback fails, ensure `afplay` is available on macOS.
+
+## License
+
+This project does not include a license. Add one if needed.

V1.7.1/main.py

@@ -0,0 +1,200 @@
+# All features in V1.7 plus improved AI models.
+import ollama
+import speech_recognition as sr
+import edge_tts
+import asyncio
+import tempfile
+from ollama import chat
+import ddgs
+import os
+import cv2
+import re
+import shutil
+import time
+from rich.console import Console
+from rich.text import Text
+
+console = Console()
+terminal_width = os.get_terminal_size().columns
+columns, _ = shutil.get_terminal_size()
+
+voice_selector = input("Choose a voice (Male/Female): ").lower().strip()
+if voice_selector == "male":
+    VOICE = "en-US-SteffanNeural"
+else:
+    VOICE = "en-US-AvaNeural"
+
+ANSI_BOLD = "\033[1m"
+ANSI_ITALIC = "\033[3m"
+ANSI_RESET = "\033[0m"
+
+def format_for_terminal(text):
+    if text is None:
+        return ""
+    text = re.sub(r'\*\*(.*?)\*\*', f"{ANSI_BOLD}\\1{ANSI_RESET}", text)
+    text = re.sub(r'\*(.*?)\*', f"{ANSI_ITALIC}\\1{ANSI_RESET}", text)
+    return text
+def clean_for_speech(text):
+    if text is None:
+        return ""
+    return re.sub(r'\*\*|\*', '', text)
+
+async def speak_async(text):
+    clean_text = clean_for_speech(text)
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as fp:
+        temp_path = fp.name
+    communicate = edge_tts.Communicate(clean_text, VOICE)
+    await communicate.save(temp_path)
+    os.system(f'afplay "{temp_path}"')
+    os.remove(temp_path)
+def speak(text):
+    asyncio.run(speak_async(text))
+
+answer = []
+is_app_open = 0
+is_recognised = 0
+r = sr.Recognizer()
+
+while True:
+    is_app_open = 0
+    is_recognised = 0
+
+    with sr.Microphone() as source:
+        r.adjust_for_ambient_noise(source, duration=0.2)
+        Ask_anything = "*Ask Me Anything...*"
+        speak(Ask_anything)
+        console.print(Text.from_ansi(format_for_terminal(Ask_anything)))
+        audio = r.listen(source)
+    try:
+        text = r.recognize_whisper(audio, model="small.en")
+        console.print(" ", style="cornsilk1 on gray19", justify="left")
+        console.print(Text.from_ansi(format_for_terminal(text)), style="cornsilk1 on gray19", justify="left")
+        console.print(" ", style="cornsilk1 on gray19", justify="left")
+    except sr.UnknownValueError:
+        console.print(Text.from_ansi(format_for_terminal("Could not understand audio")))
+        text = ""
+
+    if "open" in text.strip(".").strip(",").strip("!").strip("?").lower():
+        parts = text.split(maxsplit=1)
+        if len(parts) > 1:
+            target = parts[1].strip()
+            console.print(Text.from_ansi(format_for_terminal(f"**Opening {target}**")))
+            app = os.popen(
+                f'mdfind \'kMDItemKind == "Application"\' | grep -i "{target}" | head -n 1'
+            ).read().strip()
+            if app:
+                os.system(f'open "{app}"')
+            else:
+                os.system(f'open "https://{target.replace(" ", "")}.com"')
+            is_app_open += 1
+
+    if text == "":
+        console.print(Text.from_ansi(format_for_terminal("No input detected. Please try again.")))
+        pass
+    else:
+        if is_app_open == 0:
+            web_simplifier = chat(
+                model='ministral-3:3b-cloud',
+                messages=[
+                    {'role': 'user','content': text},
+                    {'role': 'system','content': f'''You are a web-search query simplifier.Your job:Convert the user's message into ONE concise web search query.User message:"{text}"Rules:- Keep only important keywords- Remove filler words and stop words- No emojis- No explanations- Keep the meaning accurate- Make it optimized for a search engine- Output ONLY the final search queryFormat:latest details on <simplified topic> as of 2026'''}
+                ]
+            )
+            trigger = chat(
+                model='ministral-3:14b-cloud',
+                messages=[
+                    {'role': 'user','content': text},
+                    {'role': 'system','content': f'''You are a vision-context detector.Determine whether answering the user's query requires:- a camera image- surroundings analysis- appearance analysis- object inspection- environmental contextUser query:"{text}"Respond ONLY with:yesornoSay YES only if:- the user refers to themselves- the user refers to their surroundings- the user asks about appearance- the user asks to inspect something visible- the answer requires visual contextExamples of YES:- "How do I look?"- "What's in front of me?"- "Analyze my room"- "What is this object?"- "Does my hair look good?"Examples of NO:- news- coding- facts- explanations- web searches- math- history- general questionsBe accurate.Do not guess.Output ONLY yes or no.No punctuation.No emojis.'''}
+                ]
+            )
+            if "yes" in trigger.message.content.lower():
+                is_recognised = 1
+                console.print(Text.from_ansi(format_for_terminal("**Capturing photo...**")))
+                cam = cv2.VideoCapture(0)
+                ret, frame = cam.read()
+                if ret:
+                    photo_path = os.path.join(os.getcwd(), "instant_photo.png")
+                    cv2.imwrite(photo_path, frame)
+                    console.print(Text.from_ansi(format_for_terminal("**Photo captured successfully!**")))
+                else:
+                    console.print(Text.from_ansi(format_for_terminal("**Error: Could not access camera.**")))
+                    photo_path = None
+                cam.release()
+                if photo_path:
+                    image_analysis = ollama.chat(
+                        model='ministral-3:8b-cloud',
+                        messages=[
+                            {'role': 'system','content': text},
+                            {'role': 'user','content': f'''Analyze the image based on the user's request.User request:"{text}"Instructions:- Focus mainly on the requested subject- If no subject is specified, analyze the surroundings- Be concise but useful- Be truthful and accurate- Mention important visible details- Do not hallucinate- No emojis- No unnecessary formatting- Keep the response natural and clear''',
+                                'images': [photo_path]
+                            }
+                        ]
+                    )
+                    img_out = image_analysis.message.content
+                    console.print(Text.from_ansi(format_for_terminal(img_out)))
+                    answer.append(img_out)
+                    speak(img_out)
+            else:
+                pass
+
+            if is_recognised == 1 and is_app_open == 1:
+                pass
+            else:
+                check = web_simplifier.message.content
+                results = ddgs.DDGS().text(
+                    check,
+                    max_results=3
+                )
+                response = chat(
+                    model='gemma4:31b-cloud',
+                    messages=[
+                        {'role': 'user','content': text },
+                        {'role': 'system','content': f'''You are Churo.Personality:- helpful- professional- intelligent- concise- accurate- natural soundingRules:- Use simple language- Keep responses concise- No emojis- Do not ramble- Answer directly- Be conversational but efficientMemory context:{answer}, use this when you feel that the query lacks context. Current user query:{text}Available web search results:{results}Use the web results ONLY if the user explicitly asks for:- latest news- recent updates- current information- newest details- web searchesOtherwise answer normally without relying on web results.If image analysis was already provided,do not repeat the analysis.Simply continue the conversation naturally.Never say:"Analysis provided"Instead continue naturally and intelligently.'''}
+                    ]
+                )
+                output = response.message.content
+                formatted_output = format_for_terminal(output)
+                speech_output = clean_for_speech(output)
+                aligned = formatted_output.rjust(terminal_width)
+                console.print(" ")
+                console.print(" ", style="cornsilk1 on gray15", justify="left")
+                console.print(Text.from_ansi(aligned), style="cornsilk1 on gray15", justify="left")
+                console.print(" ", style="cornsilk1 on gray15", justify="left")
+                console.print()
+                answer.append(output)
+                speak(speech_output)
+
+            voice_recogniser = sr.Recognizer()
+            with sr.Microphone() as source1:
+                r.adjust_for_ambient_noise(source1, duration=0.2)
+                prompt_text = "*Do you want to continue the conversation? Yes or No?*"
+                console.print(Text.from_ansi(format_for_terminal(prompt_text)))
+                speak("Do you want to continue the conversation? Yes or No?")
+                time.sleep(0.01)
+                console.print(Text.from_ansi(format_for_terminal("*Listening for your response...*")))
+                audio1 = voice_recogniser.listen(source1)
+            try:
+                voice_continue_recogniser = voice_recogniser.recognize_whisper(audio1, model="small.en")
+                voice_continue_recogniser = voice_continue_recogniser.strip()
+                console.print(" ", style="cornsilk1 on gray19", justify="left")
+                console.print(Text.from_ansi(format_for_terminal(voice_continue_recogniser)), style="cornsilk1 on gray19", justify="left")
+                console.print(" ", style="cornsilk1 on gray19", justify="left")
+            except sr.UnknownValueError:
+                console.print(Text.from_ansi(format_for_terminal("Could not understand audio")))
+            print()
+
+            norm = re.sub(r'[^\w]', '', voice_continue_recogniser).lower()
+            positive_patterns = {
+                "yes","yeah","yep","absolutely","sure","continue","ok","okay","y","yea"
+            }
+            if norm == "":
+                console.print(Text.from_ansi(format_for_terminal("**No input detected.**")))
+                pass
+            elif norm not in positive_patterns:
+                speak("Bye! Please Visit Again!")
+                break
+            else:
+                continue
+
+        else:
+            continue