Here is a fascinating brief example of just how much of an improvement Large Speech Models (LSMs) offer over classical ASR systems. Here is a brief excerpt from an English-language European broadcast earlier this month, transcribed through a major commercial classical ASR solution. The transcript is effectively gibberish that makes it nearly impossible to understand what was actually said:

of time, but it's coming up after the break. i did a strategy in berlin. thanks so much for watching that the news continues at the top of the all the winds here soon. the ice cold be passed, the gun needs an expedition ventures on 2 places that no one has the climate research in the i see the dogs march 3rd on the a hello. i live in the, in a couple months or 6. this one is the smallest her mom suffice to claim the name

Here is the exact same clip transcribed through a modern LSM system (GCP's Chirp), showing just how much of a massive improvement LSMs offer for accented English:

of time, that is coming up after the break, I am Diresh Pandi in Berlin, thanks so much for watching, but the news continues at the top of the hour, so we'll see you soon, goodbye, Ice cold beauty at the end of the world, patagonia, an expedition. ventures onto glaciers that no one has dared explore before, searching for data: climate research in the ice fields, unknown patagonia, starts march 3rd on dw. hello, aralarında en akıllıolan, this one's the smallest, her mom survived the quake.