Talk locally (no ChatGPT) with your documents 😄 | PrivateGPT + Whisper + Coqui TTS

[@nmstoker]

Fantastic work! I've long hoped we'd get to the point where this kind of end to end interaction was possible locally and your video shows it in action! 👍

[@ThorstenMueller]

Hi Neil, i'm really happy hearing from you. Thanks for your great feedback 😊.

[@alx8439]

Great job. With small LLMs we're slowly moving towards fully offline voice assistants which actually have brains, not just hardcoded skills

[@ThorstenMueller]

Full ack 👍

[@DrFukuro]

Danke für deine Arbeit! Du bist einer der wenigen, die noch wirklich relevanten Content erzeugen, ganz nach dem Open Source Prinzip. Wie immer praktisch, nachvollziehbar. So kann auch der kleine Mann am technischen Fortschritt partizipieren! Top. Ich spreche Englisch zwar auch ganz gut, würde mich aber sehr freuen, wenn du mehr Videos in unserer präzisen und schönen Muttersprache produziertest. 👍

[@ThorstenMueller]

Vielen lieben Dank für deine sehr nette Rückmeldung 😊.

[@historyrepeatscubed726]

Incredible and just what I've been waiting for! Thanks so much for your efforts and for making this video. I think we all realise that, with a degree of automation scripting and such, this is going to be an extremely valuable and powerful concept in the near future - and most importantly, in my mind at least, the privacy aspect is a winner for any corporations or private persons concerned about intrusions and loss of said privacy, and data. Thanks again. Peace.

[@Vito_0912]

Although itś not the aim of the video, but You can also build a own smart „home“/respond assistant. You can use „Sphinx“ to listen for an activation word and then use whisper for the query, feeding it trough GPT (I would recommend GPT-3.5, although it costs, because of the speed) and then use coqui to say it.

[@ThorstenMueller]

Thanks for your feedback 😊. This complete process might be a good usecase for Rhasspy.

[@toykotokyoto]

Nice! Just had a thought: it would be cool if for these kinds of videos the intro included a short demo of the “final user experience”. That way the viewer knows what all the install work will lead to! Just my 2 cents:) Great application and another great video!

[@ThorstenMueller]

Guude Josh 👋, thanks for your great suggestion. I'll try to include a short demo already during intro and not just at the end in future tutorials 😊.

[@Niall96]

I'm not sure if theres is requirements.txt anymore in recent versions of PrivateGPT, i'd like to see an up to date tutorial :D

[@katokira260]

Thanks ... your videos help me a lot with my project

[@ThorstenMueller]

You're very welcome. Happy to hear that 😊.

[@llupiy]

Hi Thorsten, thanks for your work. A paid implementation of mic input -> Whisper -> OpenAIChatGPT -> ElevenLabs (TTS) -> speaker output takes about 2 seconds to generate a response. Have you tried the implementation of your video on a faster computer? If so, how long did the whole process take to respond?

[@ThorstenMueller]

Thanks for your feedback and that interesting use case. As i don't have a faster computer - no 😆.

[@MeinDeutschkurs]

It‘s a miracle for me. I‘m not able to bring TTS to work on M1. I did a workaround by using „say“ command on Apple M1. Unfortunately, your voice is not available in „say“.

[@ThorstenMueller]

Do you get an error/exception on running TTS? At which step is it failing?

[@MeinDeutschkurs]

@@ThorstenMueller , yeah. We already talked about at the video on how to get it work on Apple. I struggled to run tts and tts-server. No worries, I hope to get it to work on m2 ultra. My m1 mb air is a real mess, have to reset it entirely.

[@ThorstenMueller]

@@MeinDeutschkurs Okay, good luck with a new (or cleaned) setup 😊.

[@RobinMadar]

Hello Iam stuck at the start, your link to imartinez repo is linked to different one and that repo doesn't have "requirements.txt" what I can do ? Thanks for your help

[@xxy1950]

Hello, a really exciting project with a lot of potential! This raised two questions for me: 1. How to customize the model by training it to recognize and follow specific storylines? 2. Is there generally no interface for internet use for privateGPT or was it just not configured accordingly in the current project? Thanks and keep it up!

[@ThorstenMueller]

Thanks for your feedback. 1. I've no experience on customizing a PrivateGPT based model on individual needs yet, but this could be an interesting topic. 2. You can use privateGPT UI with the ChatGPT model (API key required) if that's the question? Or do you mean that privateGPT can answer up to date question based on live internet research?

[@MeinDeutschkurs]

So, what do I think: I use something similar in my WhatsApp-Assistant. It‘s a great thing. Hope to find a model which is able to process everything in proper German. Currently I have to ask gpt-3.5-turbo. Can‘t wait to get rid of it. ❤️

[@ThorstenMueller]

Yeah, an open-source proper German LLM would be great 😎.

[@RobinMadar]

Hello, I would love to use your example as a base for my project to final work. My question it how to connect STT to this so bot gets the message and read answer shortly after. Is it possible with your approach ?

[@ThorstenMueller]

I've not dived deeper into STT in this context so i'm not sure how this will work.

[@ddsdheelitirkasiyaasaddade9793]

Great tutorials.

[@ThorstenMueller]

Thanks a lot 😊. Feedback like yours always keeps me motivated.

[@manash2449]

Thanks for the informative video!!! Do you have any recommendation on how to improve the response from the GPT4all model?

[@ThorstenMueller]

Good question. But no i've no idea how (or if) response performance can be tuned, apart from adjusting hardware.

[@katokira260]

can you make a tuto about ChatterBot

[@ThorstenMueller]

Thanks for that suggestion 😊. I'll check ChatterBot as soon as possible.

[@mahmoudabusaqer4585]

I am doing the same thing but using Unity 😁

[@ThorstenMueller]

Sounding interesting. Is this a public visible project?

[@mahmoudabusaqer4585]

@@ThorstenMueller Yes it will be public, still working on it

[@historyrepeatscubed726]

@@mahmoudabusaqer4585 - Great to hear! I'm sure, like me, we're all looking forward to seeing and hearing more from you about your project in the near future. Best of luck and every success. Peace.

[@torstenjaeckel5414]

Gude zrück

[@ThorstenMueller]

Danke 😊.

[@satvikmehta8481]

File "\privateGPT.py", line 83, in main() File "privateGPT.py", line 38, in main llm = GPT4All(model=model_path, n_ctx=model_n_ctx, backend='gptj', n_batch=model_n_batch, callbacks=callbacks, verbose=False) File "\Desktop\TEST\Privategptvenv\lib\site-packages\langchain\load\serializable.py", line 74, in init super()._init_(**kwargs) File "pydantic\main.py", line 341, in pydantic.main.BaseModel._init_ pydantic.error_wrappers.ValidationError: 1 validation error for GPT4All n_ctx extra fields not permitted (type=value_error.extra)File "privateGPT.py", line 83, in main() File "privateGPT.py", line 38, in main llm = GPT4All(model=model_path, n_ctx=model_n_ctx, backend='gptj', n_batch=model_n_batch, callbacks=callbacks, verbose=False) File "Desktop\TEST\Privategptvenv\lib\site-packages\langchain\load\serializable.py", line 74, in init super()._init_(**kwargs) File "pydantic\main.py", line 341, in pydantic.main.BaseModel._init_ pydantic.error_wrappers.ValidationError: 1 validation error for GPT4All n_ctx extra fields not permitted (type=value_error.extra) how to resolve this query

[@ThorstenMueller]

Hello, i've not seen this issue before. Maybe you can get faster and better help by opening an issue on their repo: github.com/imartinez/privateGPT/issues. Good luck 😊