https://www.youtube.com/watch?v=guW-vdBLxtg
It can be integrated with LLM in LM Studio or OpenAI endpoint format.
Built using Go and Wails, released for Windows, macOS (Universal2), and Ubuntu.
Although it was completed early yesterday morning, there may be incomplete aspects.
https://github.com/DINKIssTyle/DINKIssTyle-Local-LLM-Gateway-with-TTS
https://github.com/DINKIssTyle/DINKIssTyle-Local-LLM-Gateway-with-TTS/releases
Beta 2
- Maximizing TTS optimization.
- Now starting TTS generation during LLM responses.
Beta 3
- TTS Quality Variety
- WAV (Original): Play as generated without MP3 conversion resources for local use
- MP3 (High Quality - 128kbps): High quality overflowing with voice, but with about 1/10 the capacity of WAV
- MP3 (Medium - 96kbps): Well-balanced audio quality
- MP3 (Low Bandwidth - 32kbps): For mobile devices on slow wireless networks
- Fixed TTS playback issues in mobile Safari (WebKit)
- Fixed voice generation issues caused by certain special characters
- Fixed speech bubble speaker button malfunction
Beta 4
- More optimization for TTS generation and playback
- User settings added
- -- You can specify the TTS generation inference steps. Default is 5,
- Even with 2 steps, it has much better quality compared to classical TTS. Try 8 steps as well.
- -- You can adjust the CPU threads used for TTS generation inference.
- Raising it above 2 will be slower due to bottlenecks. Currently...
- Shows a list of LLM servers so you don't have to manually enter the LLM model key (model loads on first conversation)
- Working to prevent mobile device screen from turning off during TTS audio playback.
Beta 5
Added startup on boot, minimize to tray, and auto-start server options
On Windows and Linux, you can close the window and reside in the System Tray.
On macOS, you can close the window and reside in the Menubar (Syst2mUI) but it remains in the Dock.
Users can pre-configure System Prompt Presets.
Windows and Linux use system_prompt.json in the folder where the executable is located,
For macOS, ~/Documents/DKST LLM Chat/system_prompt.json is used.
Users can use customized pronunciation dictionaries.
Windows and Linux use dictionary_language.txt in the folder where the executable is located,
For macOS, ~/Documents/DKST LLM Chat/dictionary_language.txt is used.
It is used according to the TTS language in the web chat window settings.
Usage example: macOS, MacOS ; The word macOS is pronounced as MacOS.
There is a dictionary editor (Dictionary_editor.py) GUI tool in early development stages.
When accessing web chat for the first time, it checks the usage environment and is preparing to provide troubleshooting guidance.
Sound file size and time taken to send to client have been added to the console log
