← sources

Best ASR & TTS models to date to use locally

source: reddit (reddit:LocalLLaMA)published 2026-11-03 08:18ingested 2026-11-04 02:41

retained text

TL;DR  
Best Model for ASR ? So far for me Qwen3-ASR > Whisper > Kroko  
Best Model for TTS ? KokoroTTS > Piper ?!  
Any suggestions/recommendations ?!

I was truly surprised at how low info and benchmarks are available in these areas compared to LLMs even artificial analysis isn’t clear enough 

You could pick an LLM check active parameters and cost and latency and speed and quality and openness and everything 

Here it is way more complex 

Hi everyone

I hope that you are all well.

Kindly I was building an App that needs good TTS & ASR engines, ideally low latency can run in the browser and maybe streaming, it can also run locally on a server as an option if the latency is good enough, multilingual is a plus (English Arabic German French and Spanish mainly)

Initially,  
For TTS piper surpassingly worked well (my tasks usually pronounce one word) but Kokoro server side for longer sentences

And for ASR I used whisper but it was too slow but not server side and I used kroko in browser for low latency passes (then whisper server side would correct it)

Recently I found out that despite being nearly as big as Whisper Large or Turbo Qwen3-ASR works just fine and even supports streaming natively and is quickly server side and have one of the lowest error rates

I am going to use Qwen3-TTS too give it a try

Is there anything else I should use you would recommend ?!

document fetched 2026-11-04 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1xmnepj/best_asr_tts_models_to_date_to_use_locally