Download AI Chat 1.44 — Faster Voices and Speech Recognition — Reconnect Button

The AI Chat password changed on February 3rd, 2025. Current subscribers can get a new password in the welcome note in the membership tab at https://www.patreon.com/vamx/membership

New Patrons Only: Download vamX for AI Chat

Upgrade AI Chat: Download vamX.1.44.var for Virt-a-Mate 1.22 Creator (or from Mega).

Automatic Reconnection
The server now automatically reconnects if the audio connection is lost. In the past if your Internet connection was slightly unstable you could lose speech connection to the server (she wouldn't speak, and couldn't hear you speak). You would then have to press Reset AI Chat (and lose your current chat). Automatic reconnection allows you to continue your chat most of the time, but doesn't work if the server crashed and was restarted (in which case you do need to restart your conversation and choose to Reset your AI Chat to the USA or Europe server).

Quicker Voice Recognition (speech recognition)
This makes the overall AI response speed faster when you are talking to the AI (using your voice). The new (Deepgram) voice recognition is more accurate and quicker (than Whisper) although it still isn't perfect. If you say something and are misunderstood, press Edit Last, then try speaking (clearly) again (or get a better microphone). If what the AI says shows up also as if it was something you said, then your microphone can hear your headset speaker and you need to turn down either your speakers or your microphone sensitivity.

Added new Improved Quick Voices
To use these you need to switch to them in the voice drop down. These have better voice quality than the Quick Voices but are much faster than the Improved voices. These voices work in English only, but when used can speed things up dramatically (compared with using Improved Voices, or other non-quick voices).

Added new Quick Speech for Quick Voices option to split the text into optimal chunks for the new Improved Quick Voices. These voices prefer to broken down into larger chunks, and generate even a moderate amount of text in a fraction of a second.

Added Quick Speech (Don't Split) and Very Quick Speech (Don't Split). These help improve audio quality of voices by letting the voice generate the entire sentence / paragraph at once. This takes longer (for the audio to be generated) but can result in slightly higher quality audio for most voices. Very Quick Speech is the same as Quick Speech, but instructs the language model to generate a shorter reply.

Added X 3 personality (from the video demo) to the main personalities list.

Added the new Many Slow Unreliable Models item when choosing a NSFW model. This is a library of over 100 language models, most of them 70B (large). These models generally have a 10-15 second delay before each response starts, and if the wait is > 15 seconds vamX switches to responding with Lunaris. These models will not always work, but it can be fun or interesting to play with this huge variety of possible responses. There are also some models that should work well specifically in Japanese (although we haven't tested this) including EZO, Swallow, Himeyuri and Karakuri. If you are a more advanced user, and understand something about different LLMs, you might want to explore this area. For beginners, these models probably won't offer anything better than sticking with Lunaris, or Llama 70b.

Fixed long standing issue that allowed for duplicate text/audio pairs (you would see, or hear, some part of the conversation twice).

Fixed issue where when AI responses started with "(" or "*", and a narrator voice was selected, then the narrator voice would also continue and read other parts of the sentence. This would cause speech to be heard from some invisible global audio source, instead of appropriately coming out of the Female's mouth.

Fixed issue where having "*" in the initial dialog or elsewhere in the personality could stop the AI Chat from working (at all) when using that personality.

If you are using Seamless Animations in AI Chat, they start to play before / while she is speaking so she can animate while simultaneously speaking (if possible).

Fixed issue where audio was played stating the password was wrong (on first open of AI Chat in vamX), even if you had the correct password entered.

Fixed issue where some (bad?) mice would send additional mouse down commands when released. This caused speech to not be recognized in desktop mode (you would hold the right mouse button down, speak, then release it, but the AI wouldn't respond).

The default Narrator option is now "Remove Would-be-Narrated". That means if the AI outputs something like "Hi Jack (she smiles and waves). What's up?" then the description is removed so she would simply say "Hi Jack. What's up?". This makes the conversation feel more natural. If you want a more rich AI generated story (and less of a conversation) choose a narrator voice to hear her say "Hi Jack" and the invisible narrator voice would say "she smiles and waves", etc.

GENERAL VAMX (NON-AI CHAT) FIXES

Fixed issue where voice commands (including from the AI) could switch vamX from Seamless Animations to Standing Prop Poses (or other timeline animations), causing a huge delay or hang / crash of Virt-a-Mate. The seamless animations are large and once loaded most computers shouldn't switch from them to other timeline animations. We also now mention when loading Seamless Animations that switching to other types of animations can crash Virt-a-Mate.

Fixed issue where vamX VaMSync didn't work for non-vamX scenes where the people weren't named Person (female) and Person#2 (male).

Added Seamless Animation transition between Kneeling and Lying Down without standing up in between.

If you enable Lock Look then the stripping sequence no longer triggers (as this would change her clothes even though they should have been locked with lock look).

BUGS / TO DO

Need to fix an issue where vamX Actions that were supposed to be disabled (non-AI vamX voice commands) stay active while using AI Chat. This is important as extra actions load while you talk to the Chat AI (using the local vamX action system instead of the correct AI Chat action system that includes her consent, uses a more limited number of actions that make sense in a conversation, etc.). For now, most users should probably go to the vamX Sound tab and uncheck Speech Recognition when using AI Chat.