Latest posts
Chapi Dev Talks
22 Sept, 13:29
Got it on hand!
Chapi Dev Talks
22 Sept, 13:19
At Science and Museum.
Chapi Dev Talks
21 Sept, 21:59
using chapi's model to caption tiktok videos 👀Chapi Dev Talks
21 Sept, 21:59
Vivid Just made this and so cool to see in action1,700Open in TelegramChapi Dev Talks
21 Sept, 17:05
hohe.et👀Chapi Dev Talks
21 Sept, 13:00edited
Hohe ASR now ships with the server we run ourselves, and it is open.Three commands and you have Amharic speech to text running on your own machine. No GPU, no account, no API key, nothing to sign.git clone https://github.com/snapwre/hohe-serve && cd hohe-serve docker build -t hohe-asr . && docker run --rm -p 8080:8080 hohe-asr curl -F audio=@clip.ogg http://localhost:8080/transcribeThe image fetches the weights from Hugging Face itself and serves on ordinary processor cores. On a clean machine we timed it end to end: built from nothing, model loaded in 6.8 seconds, a 6 second Amharic clip transcribed in 1.1.It streams too. Long audio is cut at its quietest moments and the text comes back in pieces as it works, rather than all at the end.Also in the repository: exactly how we deploy it on AWS spot behind a load balancer, and the full speed measurements behind every number weGitHubGitHub - snapwre/hohe-serve: Hohe ASR as a service: Amharic speech to text on ordinary processors, no GPUHohe ASR as a service: Amharic speech to text on ordinary processors, no GPU - snapwre/hohe-serveChapi Dev Talks
21 Sept, 12:03
Laya on Mac M4 CoreML Offline (Score: 150+ in 15 hours) Link: https://readhacker.news/s/75sNU Comments: https://readhacker.news/c/75sNU
Chapi Dev Talks
21 Sept, 10:13edited
The Fact that hohe ASR is running on 4VCPU is insane honestly and so small parameter 0.6B and it performs very well.anytime try it on https://dataset.et/ai or @dataset_ai_botAND OPEN SOURCE 🤯dataset.etTry Hohe, Amharic speech to textSpeak Amharic and watch it written, live in your browser.Chapi Dev Talks
21 Sept, 08:09
forwarded from @hacker_news_feed
Laya on Mac M4 CoreML Offline (Score: 150+ in 15 hours)Link: https://readhacker.news/s/75sNU Comments: https://readhacker.news/c/75sNUGistLaya on Mac m4 CoreML Offline https://github.com/mizorewww/laya-coremlLaya on Mac m4 CoreML Offline https://github.com/mizorewww/laya-coreml - laya.sh1,880Open in TelegramChapi Dev Talks
21 Sept, 04:51
Good Morning fellasHave a great week ahead!Try Hohe ASR: https://dataset.et/aiWhat a great way to start the daydataset.etTry Hohe, Amharic speech to textSpeak Amharic and watch it written, live in your browser.Chapi Dev Talks
20 Sept, 22:27
Good Night enough for today!https://t.me/chapidevtalks/3038TelegramChapi Dev TalksHohe ASR is out. Open Amharic speech to text. Start with the speed, because it is the part people do not expect. On four ordinary CPU cores, a five second voice note is transcribed in 0.39 seconds. A thirty second clip takes 7.75 seconds. An hour of audio…Chapi Dev Talks
20 Sept, 22:27edited
Hohe ASR is out. Open Amharic speech to text. Start with the speed, because it is the part people do not expect. On four ordinary CPU cores, a five second voice note is transcribed in 0.39 seconds. A thirty second clip takes 7.75 seconds. An hour of audioChapi Dev Talks
20 Sept, 21:55
support us by sharing the messagehttps://x.com/chapimenge3/status/2101789761031557562?s=20https://lnkd.in/p/epe26bNiX (formerly Twitter)Chapi Menge (@chapimenge3) on XHohe ASR is out: open Amharic speech to text, built in Addis Ababa. A five second voice note transcribed in 0.39 seconds. On a CPU. No GPU anywhere. 16.1% word error on speakers it has never hea…Chapi Dev Talks
20 Sept, 21:47
Hohe ASR is out. Open Amharic speech to text. Start with the speed, because it is the part people do not expect. On four ordinary CPU cores, a five second voice note is transcribed in 0.39 seconds. A thirty second clip takes 7.75 seconds. An hour of audio1,690Open in TelegramChapi Dev Talks
20 Sept, 21:42
Hohe ASR is out. Open Amharic speech to text.Start with the speed, because it is the part people do not expect. On four ordinary CPU cores, a five second voice note is transcribed in 0.39 seconds. A thirty second clip takes 7.75 seconds. An hour of audio goes through in under ten minutes, on a machine that costs about nine cents an hour to rent. No GPU anywhere in that sentence.It is a CTC model, one single pass over the audio instead of generating word by word, which makes it roughly five times faster than a Whisper of similar accuracy. It also means it does not invent sentences out of silence.Accuracy: 16.1% word error and 5.4% character error on speakers it has never heard. 880 hours of training audio, including five regional dialects and a third of it pushed through a phone line on purpose.What it is bad at, plainly: two people talking at once, around 47% word error. Numbershuggingface.cosnapwre/hohe-asr-amharic · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.Chapi Dev Talks
20 Sept, 21:37
New Open source model releasing tonightChapi Dev Talks
20 Sept, 21:17edited
software guy on seyfu 🔥 @software_guy
Chapi Dev Talks
20 Sept, 19:15
new home
Chapi Dev Talks
20 Sept, 19:12
Sad to see let it go!i will be back soon untill then adios!
2,040Open in TelegramChapi Dev Talks
20 Sept, 09:44
So the past couple of weeks quietly I was training a model that can have a CER (character error rate) and WER (word error rate) for Amharic speech to text models and pretty much there is so many out there but not quite good one so far. So our new open source
Related Channels
Other channels in the same section of the catalogue.
