SakeTami
vamx
vamx

patreon


vamX Deepgram App - Separate Speech Recognition App - Faster and More Accurate + Custom Personalities

SEPARATE SPEECH RECOGNITION APP

This app gives the whole vamX Chat AI a more conversational feel.

You don't hold down buttons to speak to her, just talk naturally.

This makes voice recognition much faster, which makes her respond much faster. Your speech is transcribed as you speak, instead of waiting for you to finish speaking. This can mean the Chat AI knows what you said almost immediately after you finished speaking, and you may receive an immediate response.

You need to download a separate app that we made, so you can have your voice sent directly to speech to text servers, without passing through Virt-a-Mate, for much faster results.

Speech recognition requires a high quality headset with a boom microphone (a microphone that is placed near your mouth), or the built-in microphone from a good VR headset. Background noise can be translated to random text by the AI.

This app also allows you to create a list of custom personalities (if you want to create your own personalities and store them on your computer). See the HOW_TO_USE_DEEPGRAM.txt file.

Download the vamX Deepgram Speech Recognition app:
https://vamx.b-cdn.net/DeepgramApp.zip
(Updated March 29th, 2025)


Download DeepgramApp.zip, unzip it, then read the instructions in the HOW_TO_USE_DEEPGRAM.txt file. This also explains how to save custom personalities using this app.


DeepgramApp is clean, but Avast & AVG flag it, as seen in this Virus Total Scan (see the rest of the antivirus checks). You can also upload the .exe to virustotal.com yourself to scan it.
https://www.virustotal.com/gui/file/bb18a657cdbb49225017f8531ac1067edbab29c55bac8e4c314b5515d4128f71?nocache=1

Added March 24th, 2025: Continuous Speech Recognition with End-of-Speech Keyword.
Say any selected word to finish speaking and get an AI response. The used keyword is removed, so if "banana" is your keyword, and you say "Let's go play banana" the AI responds to "Let's go play". This makes Deepgram very quick to respond, and also wait to respond until you are finished talking (so you don't get interrupted while talking, if you talk slowly).

Added June 22nd, 2024: There are now many additional languages recognized, ability to set the delay before your speech is considered finished, and even the ability to use the Deepgram app to create a private personality library on your computer. This speech recognition should also be more accurate.

Please note, if you start seeing what she says appear as if you said it, your speaker/headset is too loud (loud enough that your microphone hears the speech from your headset). If your microphone hears any speech, it will be transcribed!

Comments

AI voice recognition is different than standard voice recognition. It, like Chat AI, often "hallucinates". By this I mean that if you aren't very careful, it prefers to guess that silence is some random set of words, instead of silence. This is also true with Whisper. In fact, when we try to get results from whisper, I send a silly AI instruction string something like "The following voice sample probably is silence." otherwise it hallucinates too much. So don't worry if it generates random stuff, AIs just currently tend to that.

vamX

na...i ment i cant insert an image into the comment section here. no, no TV running or else and even if, it would have shown in deepgrams interface wouldnt it? using the builtin laptop microphone and never had issues with that before. again if deepgram would have picked something up is would have written it in its interface wouldnt it? that wasnt the case so that suggests it must have been injected from outside of my pc?

AgentXXDC

Insert images? Explain more. Where do you want to insert them? I guess you mean that deepgram thinks that's what you said? Did you have the TV on in the background??? Perhaps check your microphone settings in Windows, maybe the microphone volume is set very low? Are you using a headset microphone? Built-in computer micrphone's are generally not good enough.

vamX

mhh cant insert images... anyway, just tested the deepgram tool and its actually sending advertise. "Got to Beadaholique.com for all your beading supply needs!" Its nothing i said or noise that it made up coz that wasnt sent from the deepgram app on my PC. also current model loaded was sfw chatgpt 4.

AgentXXDC

If that happens you need to turn down the volume of your headset / speaker, as your microphone is hearing her talk.

vamX

Curious George


More Creators