OpenAI releases Whisper v3, new generation open source ASR model

Good improvements for many languages, numbers here<p><a href="https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;whisper&#x2F;blob&#x2F;main&#x2F;language-breakdown.svg">https:&#x2F;&#x2F;github.com&#x2F;openai&#x2F;whisper&#x2F;blob&#x2F;main&#x2F;language-breakdo...</a>

Related ongoing threads:<p><i>New models and developer products</i> - <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38166420">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38166420</a><p><i>OpenAI DevDay, Opening Keynote Livestream [video]</i> - <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38165090">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38165090</a>

Does anyone know of a nice UI wrapper for something like whisper.cpp?<p>I need to write a lot of long texts for work and some good dictation software would be great. I know there&#x27;s Dragon, but somehow I have not been able to find something that fits my need and is free.

This seems like the best free voice recognition in general.<p>Is there a model that is the best at wake word detection? The last that I looked, it seemed like this was fairly lacking.

Still doesn&#x27;t look like it can do real-time unfortunately.<p>Edit: I understand that you can use small samples and approximate something like streaming, but the limitation here is you wind up without context for the samples, increasing WER. It would be nice if there was some streaming option.

This is great, but I hope in the future there would be a speech-to-text model with a focus on low-resource languages, probably by balancing the dataset similar to No Language Left Behind (NLLB) released by Meta, it&#x27;s a translation model that works really well even with low-resource languages, it would be really cool something similar for speech transcription.

They say whisper-3 will be available via the api soon. Does anyone know why only whisper-1 was ever made available via the api (no whisper-2)?

Only 3GB, interesting to see how small SOTA models in other domains are compared to LLMs like Falcon-180B.

did they break the api?<p>from openai import OpenAI<p>Traceback (most recent call last): File &quot;&lt;stdin&gt;&quot;, line 1, in &lt;module&gt; ImportError: cannot import name &#x27;OpenAI&#x27; from &#x27;openai&#x27;<p>If so where is the current documentation?

Does anyone know if it’s able to do diarization with 3?

With comments GitHub looks like HN except one less click to click.

Word from my GenAI contact is that this (or similar announcement) replaces the need for RAG.