As a linguist, I would like to know more about the kinds of languages this works well with, or does not work well with. For example, half the world's languages are tone languages, and the way tones work varies greatly among these. Some just have high and low tones, while others are considerably more complicated; Thai has high, mid, low, rising and falling. Also, tone is relative, e.g. a man's high tone might be a woman's low tone. And some African languages have tones whose absolute frequencies vary across an utterance. So transcribing tone is a quite different problem from transcribing phonemes--and yet for many tone languages, the tone is crucial.
There are also rare(r) phonemes, like the clicks in many languages of southern Africa. Of course maybe they've already trained on some of these languages.
The HuggingFace demo says "Supported Languages[:] For this public demo, we've restricted transcription to low-resource languages with error rates below 10%." That's unclear: 10% word error rate, or character/ phoneme error rate? The meta.com page refers to character error rate (CER); a 10% character error rate can imply a much higher word error rate (WER), since most words contain several characters/ phonemes. That said, there are ways to get around that, like using a dictionary to select among different paths through possible character sequences so you only get known words, and adding to that a morphological parser for languages that have lots of affixes (meaning not all the word forms will be in the dictionary--think walk, walks, walked, walking--only the first will be in most dictionaries.)
Enquiring minds want to know!
South Estonian: "vulnerable" – sure, yeah
Karelian: "endangered" – seems correct
Swedish: also "endangered" – wat
Ghari (12k speakers): "safe" – :facepalm:
Are these really language-vulnerability ratings or did they just make a mapping from Trump's tariff rates?
Freely downloadable and usable by anyone for almost anything.
We truly live in the future.
Music is the universal language, but one day soon it will be replaced by Chinese.
> Omnilingual ASR was designed as a community-driven framework. People around the world can extend Omnilingual ASR to new languages by using just a few of their own samples.
The world just got smaller
Half joking - hopefully, we can still contribute something to this to this field. Looking forward to doing some tests with this.
Also, 1.6k < 6k, and I highly doubt this model is anywhere near as good as it is on EU languages for most of them.
And their use of LLMs as part of the transcription process makes it likely that they trained the model to correct mispronounciations by the speaker. This lowers CER because the human transcription often corrects for mispronounciations as well, but reduces the ability of the model to actually transcribe what was said.
Original post:
You can use the OmniASR SSL models instead of their older MMS models to create TTS models: https://github.com/ylacombe/finetune-hf-vits
What might be interesting is the newly released OmniASR data, because the MMS data, which was used for the MMS TTS, was never released.
Also, the OmniASR can be used to transcribe some untranscribed speech to train a TTS on it.
[1] MMS paper: https://arxiv.org/pdf/2305.13516
What would it take to start working on them?
"You haven't experienced Shakespeare until you've read him in the original Bonobo". :-)
Status: Endangered
"The child-bearing generation can use the language among themselves, but it is seldom being transmitted to children."
What!? A lot must have changed in one generation..
Few-shot new languages is going to be a game changer for linguists
https://aidemos.atmeta.com/omnilingualasr/language-globe
- we are getting closer to BabelFish.. at least for the Earth!
can't say for sure, but a lot of the UI (and text) is quite familiar. the history page is a near rip off which is a giveaway.
i believe the mit license should be distributed since it's almost certainly a derivative work.
"The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software."
I can't confirm if the license is infact distributed, since I would have to pay $50, which quite frankly I'm not going to do.
a bit sad to see a ui reskin claimed as original work. the reskin is totally fine, but I believe the license must be distributed. i believe in the proliferation of this software so im happy to see this overall (it's good enough someone wants to charge for it! that's a big win!) but it's just a bit of a shame how this project has gone about it imo.
I am trying to approach this will full transparency, honesty and respect for what the creator of Handy did. I'm not a grifter.
Please consider that my project is still very young. I didn't include the third-party licenses in my first few releases (I honestly didn't know this about the MIT license, my bad!), but will fix this asap with the next release (hopefully coming out in a few days), and I'll pull the previous releases to avoid distributing versions that don't include the licenses. I'll also add information about the other production dependencies that I'm using.
If you look at my announcement, you'll see that I'm being fully transparent about this and am not interested in cloning Handy at all. My code is already very distant from the initial version I started with and I'm exploring and building features that will probably never be included in Handy. For instance, my app's UI has been created from scratch (with a lot of inspiration from Handy), it is fully responsive and now works on Omarchy (Hyprland/Wayland), which Handy doesn't support at the moment. I have added various features for my own needs and for my first customers (e.g., . In the roadmap of the product, you can see some of the ideas I intend to develop.
I also intend to contribute back to Handy over time. I already have and will continue to do so.
As for "very clean data," see section 5.7.4: "Omnilingual + OMSF ASR was intentionally curated to represent naturalistic (i.e., often noisy) audio conditions, diverse speaker identities, and spontaneous, expressive speech."
The core of the transcription process is mostly the same and uses the same libraries, but I've rebuilt the UI from scratch to be fully responsive, added support for Wayland/Hyprland (with adaptive window size), implemented lazy loading/filtering/group by date/editing/search/filtering etc to the history screen, implemented history storage/handling differently, added more control over the history feature, added support for custom sounds, improved UX around managing sounds, added a loading screen, added support for model download pause/resume/cancel/delete etc.
These might seem like details, but it all takes time. I started this project 3 weeks ago and this is just the beginning.
In my roadmap I've listed many ideas I have in mind and will be focusing on: https://docs.voice-ai.knowii.net/roadmap
I want to go in a different direction than Handy, and my customers (who are mainly interested in Knowledge Management) too.