Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
This is an on-CPU real-time conversational system for two-way speech communication with AI models, utilizing a continuous streaming architecture for fluid conversations with immediate responses and natural interruption handling.
| Date | Stars |
|---|---|
| 2026-07-24 | 255 |
| 2026-07-25 | 255 |
| 2026-07-28 | 255 |
| 2026-07-30 | 255 |
| 2026-08-06 | 255 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
[](https://youtu.be/x92FLnwf-nA) [](https://www.linkedin.com/in/abdullahalasif-bd/) [](https://github.com/asiff00/On-Device-Speech-to-Speech-Conversational-AI/stargazers) [](https://github.com/asiff00/On-Device-Speech-to-Speech-Conversational-AI/network/members) # On Device Speech to Speech Conversational AI [](https://github.com/asiff00/On-Device-Speech-to-Speech-Conversational-AI/raw/main/assets/video_demo.mov) This is a real-time conversational system for two-way speech communication with AI models, utilizing a continuous streaming architecture for fluid conversations with immediate responses and natural interruption handling. All components of this system are run locally [on CPU, in my test system]. <h2 style="color: yellow;">HOW TO RUN IT</h2> 1. **Prerequisites:** - Install Python 3.8+ (tested with 3.12) - Install [eSpeak NG](https://github.com/espeak-ng/espeak-ng/releases/tag/1.52.0) / `sudo apt install -y espeak-ng` for Linux (required for voice synthesis) [Linux user check this issue](https://github.com/asiff00/On-Device-Speech-to-Speech-Conversational-AI/issues/7#issuecomment-2661541707) - Install Ollama from https://ollama.ai/ 2. **Setup:** - Clone this repository `git clone https://github.com/asiff00/On-Device-Speech-to-Speech-Conversational-AI.git` - Run `cd On-Device-Speech-to-Speech-Conversational-AI` and go to the project directory - Run `git lfs pull` from the project root directory to download the models and voices - Copy `.env.template` to `.env` - Add your HuggingFace token to `.env` - Install requirements: `pip install -r requirements.txt` - Add any missing packages if not already installed `pip install <package_name>` 4. **Run Ollama:** - Start Ollama service - Run: `ollama run qwen2.5:0.5b-instruct-q8_0` or any other model of your choice 5. **Start Application:** - Run: `python speech_to_speech.py` - Wait for initialization (models loading) - Start talking when you see "Voice Chat Bot Ready" - Long press `Ctrl+C` to stop the application </details> # How does it work? We basically put a few models together to work in a multi-threaded architecture, where each component operates independently but is integrated through a queue management system to ensure performance and responsiveness. The flow works as follows: ### Loop (VAD -> Whisper -> LM -> TextChunker -> TTS) To achieve that we use: - **Voice Activity Detection**: Pyannote:pyannote/segmentation-3.0 - **Speech Recognition**: Whisper:whisper-tiny.en (OpenAI) - **Language Model**: LM Studio/Ollama with qwen2.5:0.5b-instruct-q8_0 - **Voice Synthesis**: Kokoro:hexgrad/Kokoro-82M We use custom text processing and queues to manage data, with separate queues for text and audio. This setup allows the system to handle heavy tasks without slowing down. We also use an interrupt mechanism allowing the user to interrupt the AI at any time. This makes the conversation feel more natural and responsive rather than just a generic TTS engine. ## Demo Video: A demo video is uploaded here. Either click on the thumbnail or click on the YouTube link: [https://youtu.be/x92FLnwf-nA](https://youtu.be/x92FLnwf-nA). [](https://youtu.be/x92FLnwf-nA) ## Performance:  I ran this test on an AMD Ryzen 5600
Excerpt of 9,912 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:b30f2a7fcd50cf01, topic:tts, topic:asr, topic:audio-processing