KoMETA AI models
This is a collection of models trained on KoMETA's talents' voice, manner of speech, and other things.
Model Description (SVC)
<!-- Provide a longer summary of what this model is. -->
- Developed by: Cooper "Elektriksan" P.
- Model type: Voice Conversion (so-vits-svc)
- License: CreativeML OpenRAIL-M
Dataset
ElaineGeneral1
Around 3500 audio files of Elaine's voice, totaling to more than 3 hours worth of audio.
Singlaine2
130 audio files of Elaine's unarchived karaoke streams, singing voice only.
Model Description (Text Generation)
- Developed by: Cooper "Elektriksan" P.
- Model type: Low Rank Adaptation (LoRA)
- License: CreativeML OpenRAIL-M
Known compatible models
7B LoRA
- OPT 6.7B
- LLaMA 7B
- GPT-4chan
- Might work with other models with shape 4096 (hidden size 4096)
1.3B LoRA
- OPT 1.3B
- Might work with other models with shape 2048 (hidden size 2048)
Dataset
VirgilCorpus
A collection of text transcribed by OpenAI Whisper (Medium). Using 13 livestreams from Virgil's channel.
- VirgilCorpusLarge (llama7b-ultra-v1, opt1.3b-ultra) (630 KB)
- VirgilCorpusBase (opt1.3b) (323 KB)
- VirgilCorpusSmallFix (llama-7b-a, llama-7b-b) (144 KB)
- VirgilCorpusMini (llama-7b) (40 KB)
VirgilCorpusV2
A better filtered and larger collection of text, using transcription by OpenAI Whisper (Medium and Small.en). Using VirgilCorpus with an additional 13 streams.
Intended Use
For entertainment, educational, and personal use only.
Out-of-Scope Use
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
Please note that these AI models are not designed for any harmful, malicious, or deceptive activities, and it is the users' responsibility to make sure that these are not used for such purposes.
Limitations and Biases
So-vits-svc (voice conversion) models can't perfectly imitate the voice of the character, this is mostly due to badly filtered data.
Text generation models have a tendency to hallucinate and spew inaccurate information or derail from the current conversation.
How to Get Started with the Model
Use so-vits-svc-fork for so-vits-svc. Use text-generation-webui for text generation. Models can be found on Huggingface.
