tevr-asr-tool
AI
speech recognition
SHOW HN2022
231points
128comments
#224of 2,570 in 2022
Top 9%of 2022 by points
How HN took it
MixedNo clear lean
9 positive, 20 neutral or asking, 5 mixed, 23 critical, out of 57 comments about the project. Thread score -0.29.
Idea: 3 for, 13 againstExecution: 1 for, 5 againstPerformance: 2 for, 2 againstDocs: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Impressive compact implementation compared to DeepSpeech
From 2 comments
If you compare this with Mozilla's DeepSpeech repository, you will find that correctly calling TensorFlow Lite and handling the results in only 284 lines of code is impressive,…
fxtentacle on HNTo those criticizing the title, how would you improve it? Dependencies do the heavy lifting , but that's true even for a hello-world. It seems obvious enough that I wouldn't see…
lynndotpy on HN
Readable source code
From 1 comment
I think the author should probably use some C++ <algorithm>'s in lines 241 and later, avoiding raw loops and making the code clearer. Also, yeah, it's not 284 lines of C++ by any…
einpoklum on HN
Objections
284 line count ignores large dependencies and model weights
From 4 comments
And 27 includes plus a huge model. "284 lines of C++" is something you could fit on a small microcontroller. This isn't 284 lines of C++.
postalrat on HN284 lines of C++ and how many trained weights? Touting the "low number of lines" on a neural network project seems kind of silly, since the logic is encoded in the weights. Kind…
woah on HNI don’t want to take anything away from the achievement, because this looks very useful, but the headline misled me a bit. I’m note sure whether it’s my affinity for code golf,…
codeflo on HNI don't know. It's not that it isn't impressive and it is very illustrative of how much you can build, utilizing the tools that are available to all of us. It's just at what…
mrweasel on HN
Possible LM overtraining inflates results
From 1 comment
This model seems strongly overtrained on CV test set. Usually improvement from LM rescoring is just 10% relative. In the paper https://arxiv.org/pdf/2206.12693.pdf the…
nshm on HN
Title misleading
From 2 comments
Includes a standard C++ library called tensorflow...
pkaye on HNI was quite interested in this, but after seeing that it includes three big other repositories, my interest was immediately gone. I'm not saying this isn't an interesting…
cgeier on HN
Requests
Streaming or real-time recognition support
From 1 comment
This is very impressive work. Could this model be capable of doing streamed input for live speech recognition?
a2128 on HN
Accuracy on accents and dialects
From 1 comment
How well does it do with accent or slight dialect influences?
BurningPenguin on HN
Accuracy on tongue-twister-like sounds
From 1 comment
I'd like to know its accuracy on the sound of Rhabarberbarbara[0] or Nackenzacken[1]. [0]https://www.youtube.com/watch?v=l3_tRPRt9x8 [1]https://www.youtube.com/watch?v=Mngm8k4_qAc
maratc on HN
Questions
How does it compare to BPE tokenization
From 1 comment
The "TEVR" token encoding idea is interesting. Any chance you compared it with other subword tokenization techniques like BPE?
gok on HN
Can it be statically linked for embedded use
From 1 comment
Is it possible to get speech recognition into a statically linked executable? ldd myprogram should say "not a dynamic executable" Are there any libraries for those of us in…
fwsgonzo on HN
Alternatives named
Mozilla DeepSpeech
From 1 comment
In my opinion, the focus here should be on the fact that this is a state of the art AI which beats Facebook's wav2vec2 by a relative 16% improvement, Scribosermo (based on NVIDIA…
fxtentacle on HN
Vosk
From 1 comment
Also in Table 6, you see that Facebook's wav2vec 2.0 XLS-R went from 12.06% without LM to 4.38% with 5-gram LM. In comparison to that, I found TEVR going from 10.10% to 3.64%…
fxtentacle on HN
wav2vec2 XLS-R
From 1 comment
Also in Table 6, you see that Facebook's wav2vec 2.0 XLS-R went from 12.06% without LM to 4.38% with 5-gram LM. In comparison to that, I found TEVR going from 10.10% to 3.64%…
fxtentacle on HN
Kaldi
From 1 comment
In my opinion, the focus here should be on the fact that this is a state of the art AI which beats Facebook's wav2vec2 by a relative 16% improvement, Scribosermo (based on NVIDIA…
fxtentacle on HN