nshm

6 ms·

1.
by nshm 19d ago
Btw, did anyone notice they removed access to "who starred the project" recently. You can only check your own repos, not external ones. It was very nice to find related people before.
2.
by nshm 3mo ago
From the sources it is based on BL606P https://en.bouffalolab.com/product/?type=detail&id=16 voice processing is in hardware unfortunately, but it exposes some things like DOA
3.
by nshm 3mo ago
You'd better use offline stt/tts for this.
4.
by nshm 4mo ago
Kroko models are more accurate and their size is just a hundred megabytes compared to parakeet (2.5 gigabytes in default fp32)
5.
by nshm 9mo ago
> Running a family was a brutal two-person job -- and the kids had to dive in to help out the second they could lift something heavier than a couple pounds. Orphanes did struggle but most families were not just two person, families were
6.
by nshm 10mo ago
You can check whale sound recognition project https://arxiv.org/abs/2104.08614
7.
by nshm 10mo ago
And moreover, you can not tune those models for practical applications. The model is originally trained on very clean data, so lower layers are also not very stable for diverse inputs. To finetune you have to update the whole model, not jus
8.
by nshm 10mo ago
This model is actually expected to be bad for popular languages, just like previous MMS it is not accurate at all, it wins by supporting something rare well but never had good ASR accuracy even for Swedish etc. It is more a research thing t
9.
by nshm 1y ago
It is useless actually. Very slow and quality is suboptimal and it is just speech generation component. See discussion here: https://github.com/SesameAILabs/csm/issues/80
10.
by nshm 2y ago
No, there are mathematical reasons LLMs are better. They are trained with multiobjective loss (coding skills, translation skills, etc) so they understand the world much better than MLM. Original post discuss that but with more words and poi
11.
by nshm 2y ago
It is actually pretty straightforward why those model "reason" or, to be more exact, can operate on a complex concepts. By processing huge amount of texts they build an internal representation where those concepts are represented
12.
by nshm 2y ago
There is also a glitch in "dialogue"
13.
by nshm 3y ago
Anyone except me thinks he doesn't look very healthy? Its strange he is kind of slow on the video where he enters the room. Maybe some biohacking.
14.
by nshm 3y ago
Yes, it is one of the important aspects. In particular if you use TTS to create an audiobook or in a video production.
15.
by nshm 3y ago
Err, I deeply respect Amazon TTS team but this paper and synthesis is..... You publish the paper in 2024 and include YourTTS in your baselines to look better. Come on! There is XTTS2 around! Voice sounds robotic and plain. Most likely a lot
16.
by nshm 3y ago
Metavoice is one of a dozen GPT-based TTS systems around starting from Tortoise. And not that great honestly. You can clearly hear "glass scratches" in their sound, it is because they trained on MP3-compressed data. There are much
17.
by nshm 3y ago
Good improvements for many languages, numbers here https://github.com/openai/whisper/blob/main/language-breakdo...
18.
by nshm 3y ago
Ok, first we screwed buffers by making them globally tracked instead of just a piece of memory. Now its time to break all binary modules again.
19.
by nshm 3y ago
Ok, but the photos look very suspicious. 1400 year gold right from the ground shouldn't shine like that. Compare to the coins here for example https://www.smithsonianmag.com/smart-news/ancient-welsh-gold...
20.
by nshm 4y ago
Great thanks a lot. So we have numbers on PTB original perplexity 8.79 quantized 9.68, already 10% worse. And PPL reported per token I suppose? Because word PPL for PTB must be around 20, not less than 10. Any numbers on more complex tasks
21.
by nshm 4y ago
Do you have the numbers? I suspect is is way worse. Original llama.cpp authors never measure any numbers as well.
22.
by nshm 4y ago
It is not really llama, it is llama quantized to 4bit. Not even the quality of original 7B. I could also quantize it to 1 bit and claim it runs on my RPI3.
23.
by nshm 4y ago
In such an actively developed area like TTS/ASR there is high chance that custom solution would fit your needs much better. The feature set of TTS is actually pretty large and hard to combine in a single ML model. No free lunch you kno
24.
by nshm 4y ago
Vosk https://alphacephei.com/vosk/lm You can restrict the vocabulary the way you like, for example, here is the chess app built with Vosk https://www.chessvis.com/
25.
by nshm 4y ago
Thats probably medium model
26.
by nshm 4y ago
Do I understand correctly banana pricing is that it costs $1.87 per hour, so the hour of audio with large model costs you about $1? Thats probably a bit too expensive compared to cloud providers.
27.
by nshm 4y ago
https://podscript.ai/ too!
28.
by nshm 4y ago
There is new better one https://github.com/rhasspy/larynx2
29.
SoundHound Raises $25M Weeks After Major Layoffs(voicebot.ai)
2 points by nshm 4y ago|0 comments
30.
by nshm 4y ago
Picovoice technology is really nice! I happily recommend it to clients looking for lightweight ASR. You might not understand but there is a huge amount of work behind this simple demo.

More ›