Morph & Beyond

We open-sourced Addis Scribe Streaming, realtime speech recognition for Amharic. It beats Google Chirp 3 on both accuracy and speed: Word error rate: 28.4% vs 32.5% First words on screen: 1.56s vs 6.67s Final text after you stop talking: 30ms vs 1,411ms On natural speech (WAXAL) we beat Google Chirp 3 in both streaming and batch mode. On FLEURS we went from 35.0% to 19.8% WER. Hohe remains the most accurate offline Amharic model in our WAXAL test, and on FLEURS the two are within a point of each other (19.8% vs 20.6%). We built Addis Scribe for the live and edge inference side, where text has to appear while people are still talking. Trained on 5,000+ hours of Amharic speech. Test sets were never used in training. Per-clip results and scripts are in the repo. Try it😊: https://docs.addisassistant.com/docs/playground/speech-to-text API and SDK docs: docs.addisassistant.com HuggingFace🤗: huggingface.co/addisai/addis-scribe-streaming @addisassistantai

The gift that keeps on giving

What’s happening?

Get the latest upstream master, branch out to work on a small fix that takes about 15 minutes, push branch, create pr, pr is 50 commits behind master and there are several conflicts. Oh the joy (Exaggerated true story)

someone should really build this.

Ah, memories, I found an old hackathon entry for gebeya we had made with some friends (from almost ten years ago). https://devpost.com/software/compra?_gl=1*1exeaal*_gcl_au*MTQ5MDU4MTQzOS4xNzkwMjMxODY5*_ga*MjA1NTE3ODQzOC4xNzkwMjMxODcy*_ga_0YHJK3Y10M*czE3OTAyMzE4NzEkbzEkZzEkdDE3OTAyMzE5MTUkajE2JGwwJGgw

Yoo

Crazy

Never supplement one part of your life with another (Financially). For example, don’t say “with the money I get from a car I rent out I will sustain a business I’m running”. If you have different hustles, isolate them. It’s called ring-fencing and I learned the hard way.

😆🤣

I would say they’re overreacting, but I’m smart enough to know I’m dumb compared to them. This is happening a little too many times for my liking.

There’s already drama that OpenAI trained their model from a solution from a researcher’s private codex session. Oh boy.

Much better marketing than “it escaped sandbox”. WOW! read more

dang

It did well compared to Sol. I see some issues here and there, but considering the prompt was a one liner, did really well. I'll try other typical models until I run out but good first impression.

I've been seeing a lot of hype regarding astra's modeling/rendering capabilities. Let's see if it can make a simple render of an Egyptian pyramid.

Hey guys, need a tip. What works best for Amharic to English translation? Any good models that perform better than google translate?

I thought they said in a few weeks. Damn it, I was hoping to touch grass today.