Having built with and tried every voice model over the last three years, real time and non-real time... this is off the charts compared to anything I've seen before.
Thank you for the link! Their playground in Mistral does not have a microphone. it just uploads files, which does not demonstrate the speed and accuracy, but the link you shared does.
I tried speaking in 2 languages at once, and it picked it up correctly. Truly impressive for real-time.
Impressive indeed. Works way better than the speech recognition I first got demo'ed in... 1998? I remember you had to "click" on the mic everytime you wanted to speak and, well, not only the transcription was bad, it was so bad that it'd try to interpret the sound of the click as a word.
It was so bad I told several people not to invest in what was back then a national tech darling:
> I tried speaking in 2 languages at once, and it picked it up correctly.
I'm a native french speaker and I tried with a very simple sentence mixing french and english:
"Pour un pistolet je prefere un red dot mais pour une carabine je prefere un ACOG" (aka "For a pistol I prefer a red dot but for a carbine I prefer an ACOG")
And instead I got this:
"Je prépare un redote, mais pour une carabine, je préfère un ACOG."
"Je prépare un redote ..." doesn't mean anything and it's not at all what I said.
I like it, it's impressive, but literally the first sentence I tried it got the first half entirely wrong.
Doesn't seem to work for me - tried in both Firefox and Chromium and I can see the waveform when I talk but the transcription just shows "Awaiting audio input".
Wow, that’s weird. I tried Bengali, but the text transcribed into Hindi!I know there are some similar words in these languages, but I used pure Bengali that is not similar to Hindi.
Well, on the linked page, it mentions "strong transcription performance in 13 languages, including [...] Hindi" but with no mention of Bengali. It probably doesn't know a lick of Bengali, and is just trying to snap your words into the closest language it does know.
I’ve been using AquaVoice for real-time transcription for a while now, and it has become a core part of my workflow. It gets everything, jargon, capitalization, everything. Now I’m looking forward to doing that with 100% local inference!
Not terrible. It missed or mixed up a lot of words when I was speaking quickly (and not enunciating very well), but it does well with normal-paced speech.
Yeah it messed up a bit for me too when I didn't enunciate well. If I speak clearly it seems to work very well even with background noise. Remember Dragon Naturally Speaking? Imagine having this back then!
In English it is pretty good. But talk to it in Polish, and suddenly it thinks you speak Russian? Ukranian? Belarus? I would understand if an American company launched this, but for a company being so proud about their European roots, I think it should have better support for major European languages.
I tried English + Polish:
> All right, I'm not really sure if transcribing this makes a lot of sense. Maybe not. A цьому nie mówisz po polsku. A цьому nie mówisz po polsku, nie po ukrańsku.
They don't claim to support Polish, but they do support Russian.
> The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch. With a 4B parameter footprint, it runs efficiently on edge devices, ensuring privacy and security for sensitive deployments.
I wonder how much having languages with the same roots (e.g. the romance languages in the list above or multiple Slavic languages) affects the parameter count and the training set. Do you need more training data to differentiate between multiple similar languages? How would swapping, for example, Hindi (fairly distinct from the other 12 supported languages) for Ukrainian and Polish (both share some roots with Russian) affect the parameter count?
Swahili is subcontinental lingua franca spoken by 200M people and growing quickly. Polish is spoken by a shrinking population in one country where English is understood anyways.
> The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch.
Yeah, it's too bad. Apparently it only performs well in certain languages: "The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch"
That's a mix of Polish and Ukrainian in the transcript. Now, if I try speaking Ukrainian, I'm getting transcript in Russian every time. That's upsetting.
Oh no! The model won't translate to an unsupported language, and incorrectly reverts to one that it was explicitly trained on.
The base likely was pretrained on days that included Polish and Ukrainian. You shouldn't be surprised to learn it doesn't perform great on languages it wasn't trained on, or perhaps had the highest share of training data.
I'm not sure why but their multilingual performance in general has usually been below average. For a French company, their models are not even close to being best in French, even outdone by the likes of Qwen. I don't think they're focusing on anything but English, the rest is just marketing.
Incroyable! Competitive (if not better) than deepgram nova-3, and much better than assembly and elevenlabs in basically all cases on our internal streaming benchmarking.
The dataset is ~100 8kHz call recordings with gnarly UK accents (which I consider to be the final boss of english language ASR). It seems like it's SOTA.
Where it does fall down seems to be the latency distribution but I'm testing against the API. Running it locally will no doubt improve that?
Is it 0.003 per minute of audio uploaded, or "compute minute"?
For example fal.ai has a Whisper API endpoint priced at "$0.00125 per compute second" which (at 10-25x realtime) is EXTREMELY cheaper than all the competitors.
I'm so amazed to find out just how close we are to the start trek voice computer.
I used to use Dragon Dictation to draft my first novel, had to learn a 'language' to tell the rudimentary engine how to recognize my speech.
And then I discovered [1] and have been using it for some basic speech recognition, amazed at what a local model can do.
But it can't transcribe any text until I finish recording a file, and then it starts work, so very slow batches in terms of feedback latency cycles.
And now you've posted this cool solution which streams audio chunks to a model in infinite small pieces, amazing, just amazing.
Now if only I can figure out how to contribute to Handy or similar to do that Speech To Text in a streaming mode, STT locally will be a solved problem for me.
Happy to answer questions about this (or work with people on further optimizing the open source inference code here). NVIDIA has more inference tooling coming, but it's also fun to hack on the PyTorch/etc stuff they've released so far.
Yeah, I think the multilingual improvements in V3 caused some kind of regression for English - I've noticed large blocks occasionally dropped as well, so reverted to v2 for my usage. Specifically nvidia/parakeet-tdt-0.6b-v2 vs nvidia/parakeet-tdt-0.6b-v3
Parakeet is really good imo too, and it's just 0.6B so it can actually run on edge devices. 4B is massive, I don't see Voxtral running realtime on an Orin or fitting on a Hailo. An Orin Nano probably can't even load it at BF16.
I noticed that this model is multilingual and understands 14 languages. For many use cases, we probably only need a single language, and the extra 13 are simply adding extra latency. I believe there will be a trend in the coming years of trimming the fat off of these jack of all trades models.
STT services that have been around for longer, like Azure, Google and Amazon, generally require you to request a specific language, and their quality is a lot higher than models that advertise themselves as LLMs (even though I believe the clouds are also using the same types of models now).
It doesn't make sense to have a language-restricted transcription model because of code switching. People aren't machines, we don't stick to our native languages without failure. Even monolingual people move in and out of their native language when using "borrowed" words/phrases. A single-language model will often fail to deal with that.
yeah, one example I run into is getting my perplexity phone assistant to play a song in spanish. I cannot for the life of me get a model to translate:
"Play señorita a mi me gusta su style on spotify" correctly
uhhh i cast doubt on multi-language support as affecting latency. model size, maybe, but what is the mechanism for making latency worse? i think of model latency as O(log(model size))… but i am open to being wrong / that being a not-good mental model / educated guess.
Even model size, it’s modest. There is a lot of machinery that is going to be common for all languages. You don’t multiply model size by 2 when you double the number of supported languages.
If encoding more learned languages and grammars and dictionaries makes the model size bigger, it will also increase latency. Try running a 1B model locally and then try to run a 500B model on the same hardware. You'll notice that latency has rather a lot to do with model size.
They've already done the inverse and trimmed non-coding abilities from their language model: https://openai.com/index/introducing-gpt-5-2-codex/. There's already precedent for creating domain-specific models.
I think it's nice to have specialized models for specific tasks that don't try to be generalists. Voxtral Transcript 2 is already extremely impressive, so imagine how much better it could be if it specialized in specific languages rather than cramming 14 languages into one model.
That said, generalist models definitely have their uses. I do want multilingual transcribing models to exist, I just also think that monolingual models could potentially achieve even better results for that specific language.
Do you have experience with that model for diarization? Does it feel accurate, and what's its realtime factor on a typical GPU? Diarization has been the biggest thorn in my side for a long time..
You can test it yourself for free on https://console.mistral.ai/build/audio/speech-to-text
I tried it on an english-speaking podcast episode, and apart from identying one host as two different speakers (but only once for a few sentences at the start), the rest was flawless from what I could see
Played with the demo a bit. It's really good at English, and detects language change on the fly. Impressive.
But whatever I tried, it could not recognise my Ukrainian and would default to Russian in absolutely ridiculous transcription. Other STT models recognise Ukrainian consistently, so I assume there is a lot of Russian in training material, and zero Ukrainian. Made me really sad.
Thats just the result of the model only supporting russian (and 12 other languages) and not urkainian. It maps to the closest words from training data.
Is there an open source Android keyboard that would support it? Everything I find is based on Whisper, which is from 2022. Ages ago given how fast AI is evolving.
I wish I had a Google Keyboard that could easily run on Whisper Medium. This is already great. But unfortunately would be too much inference cost, incredibly slow. The problem with Whisper is not the inference quality: medium and large are incredible. Is that the base model is not enough, and the only one with fast inference in mobile devices.
WER is slightly misleading, but Whisper Large v3 WER is classically around 10%, I think, and 12% with Turbo.
The thing that makes it particularly misleading is that models that do transcription to lowercase and then use inverse text normalization to restore structure and grammar end up making a very different class of mistakes than Whisper, which goes directly to final form text including punctuation and quotes and tone.
But nonetheless, they're claiming such a lower error rate than Whisper that it's almost not in the same bucket.
On the topic of things being misleading, GPT-4o transcriber is a very _different_ transcriber to Whisper. I would say not better or worse, despite characterizations such. So it is a little difficult to compare on just the numbers.
There's a reason that quite a lot of good transcribers still use V2, not V3.
3 hours for a single request sounds nice to me. Although the graph suggests that it’s not going to perform as good as openai model I have been using, it is open source and surely I will give it a try.
I really wish those offering speech-to-text models provided transcription benchmarks specific to particular fields of endeavor. I imagine performance would vary wildly when using jargon peculiar to software development, medicine, physics, and law, as compared to everyday speech. Considering that "enterprise" use is often specialized or sub-specialized, it seems like they're leaving money on Dragon's table by not catering to any of those needs.
Looks like this model doesn't do realtime diarization, what model should I use if I want that? So far I've only seen paid models do diarization well. I heard about Nvidia NeMo but haven't tried that or even where to try it out.
I was skepitcal upon hearing the figure but various sources do indeed back it up and [0] is a pretty interesting paper (old but still relevant human transcibers haven't changed in accuracy).
I think it's actually hard to verify how correct a transcription is, at scale. Curious where those error rate numbers come from, because they should test it on people actually doing their job.
I haven't quite figured out if the open weights they released on huggingface amount to being able to run the (realtime) model locally - i hope so though! For the larger model with diarization I don't think they open sourced anything.
> We've worked hand-in-hand with the vLLM team to have production-grade support for Voxtral Mini 4B Realtime 2602 with vLLM. Special thanks goes out to Joshua Deng, Yu Luo, Chen Zhang, Nick Hill, Nicolò Lucchesi, Roger Wang, and Cyrus Leung for the amazing work and help on building a production-ready audio streaming and realtime system in vLLM.
Italian represents, I believe, the most phonetically advanced human language. It has the right compromise among information density, understandability, and ability to speech much faster to compensate the redundancy. It's like if it had error correction built-in. Note that it's not just that it has the lower error rate, but is also underrepresented in most datasets.
I love seeing people from other countries share their own folk tales about what makes their countries special and unique. I've seen it up close in my country and I always cringed when I heard my fellow countrymen came up with these stories. In my adulthood I'm reassured that it happens everywhere and I find it endearing.
On the information density of languages: it is true that some languages have a more information dense textual representation. But all spoken languages convey about the same information in the same time. Which is not all that surprising, it just means that human brains have an optimal range at which they process information.
Further reading: Coupé, Christophe, et al. "Different Languages, Similar Encoding Efficiency: Comparable Information Rates across the Human Communicative Niche." Science Advances. https://doi.org/10.1126/sciadv.aaw2594
Different representations at the same bitrate may have features that make one a lot more resilient to errors. This thing about Italian, you fill find in any benchmark of vastly different AI transcribing models. You can find similar results also on the way LLMs mostly trained on English generalize usually very well with Italian. All this despite Italian accounting for marginal percentage of the training set. How do you explain that? I always cringe when people refute evidence.
This is largely due to the fact that modern Italian is a systematised language that emerged from a literary movement (whose most prominent representative is Alessandro Manzoni) to establish a uniform language for the Italian people. At the time of Italian unification in 1861, only about 2.5% of the population could speak this language.
The language itself was not invented for the purpose: it was the language spoken in Florence, than adopted by the literary movement and than selected as the national language.
It seems like the best tradeoff between information density and understandability actually comes from the deep latin roots of the language
I was honestly surprised to find it in the first place, because I assumed English to be at first place given the simpler grammar and the huge dataset available.
I agree with your belief, other languages have either lower density (e.g. German) or lower understandability (e.g. English)
English has a ton of homophones, way more sounds that differ slightly (long/short vowels), and major pronunciation differences across major "official" languages (think Australia/US/Canada/UK).
Italian has one official italian (two, if you count IT_ch, but difference is minor), doesn't pay much attention to stress and vowel length, and only has a few "confusable" sounds (gl/l, gn/n, double consonants, stuff you get wrong in primary school). Italian dialects would be a disaster tho :)
That's interesting. As a linguist, I have to say that Haskell is the most computationally advanced programming language, having the best balance of clear syntax and expressiveness. I am qualified to say this because I once used Haskell to make a web site, and I also tried C++ but I kept on getting errors.
/s obviously.
Tldr: computer scientists feel unjustifiably entitled to make scientific-sounding but meaningless pronouncements on topics outside their field of expertise.
At least some relatively well-known research finds that all languages have similar information density in terms of bits/second (~39 bits/second based on a quick search). Languages do it with different amounts of phonetic sound / syllables / words per bit and per second, but the bps comes out the same.
I don't know how widely accepted that conclusion is, what exceptions there may be, etc.
This looks great, but it's not clear to me how to use it for a practical task. I need to transcribe about 10 years worth of monthly meetings. These are government hearings with a variety of speakers. All the videos are on YouTube. What's the most practical and cost-effective way to get reasonably accurate transcripts?
In the api it took 18s to do a 20m audio file I had lying around where someone is reviewing a product.
There will, I'm sure, be ways of running this locally up and available soon (if they aren't in huggingface right now) but the API is $0.003/min. If it's something like 120 meetings (10 years of monthly ones) then it's roughly $20 if the meetings are 1hr each. Depending on whether they're 1 or 10 hours (or if they're weekly or monthly but 10 parallel sessions or something) then this might be a price you're willing to pay if you get the results back in an afternoon.
edit - their realtime model can be run with vllm, the batch model is not open
As a rule of thumb for software that I use regularly, it is very useful to consider the costs over a 10-year period in order to compare it with software that I purchase for lifetime to install at home. So that means 1,798.80 $ for the Pro version.
One week ago I was on the hunt for an open source model that can do diatization and I had to literally give up because I could not find any easy to use setup.
I don't know if that will change, but right now only the Voxtral Mini Transcribe V2 supports diarization and it's not open-weight.
The Voxtral Realtime model doesn't support diarization, but is open-weight.
does anyone know if there's any desktop tools I can use this transcription model with? e.g. something where like Wisper Flow/WillowVoice but with custom model selection
Disappointing how this lacks a clear reference implementation, if not mixed at almost yet unreleased VLLM (nightly version) stuff. I'm ok with Open Weights being a form of OSS in the case of models, because frankly I don't believe that, for large LLMs, it is feasible to release the training data, all the orchestration stuff, and so forth. But it can't be: here are the weights, we partnered with VLLM for inference. Come on. Open Weights must mean that you put me in a situation to write an implementation easily for any hardware.
p.s. even the demo uses a remote server via websocket.
Pseudo related -- am I the only one uncomfortable using my voice with AI for the concern that once it is in the training model it is forever reproducible? As a non-public person it seems like a risk vector (albeit small),
It's a real issue, but why do you only see it in ai? It's true for any case where you're speaking into a microphone
Depending on the permissions granted to apps on your mobile device, it can even be passively exfiltrated without you ever noticing - and that's ignoring the video clips people take and put online. Like your grandma uploading to Facebook a short moment from a Christmas meet or similar
There have already been successful scams - eg calls from "relatives" (AI) calling family members needing money urgently and convincing them to send the money...
This demo is really impressive: https://huggingface.co/spaces/mistralai/Voxtral-Mini-Realtim...
Don't be confused if it says "no microphone", the moment you click the record button it will request browser permission and then start working.
I spoke fast and dropped in some jargon and it got it all right - I said this and it transcribed it exactly right, WebAssembly spelling included:
> Can you tell me about RSS and Atom and the role of CSP headers in browser security, especially if you're using WebAssembly?
Having built with and tried every voice model over the last three years, real time and non-real time... this is off the charts compared to anything I've seen before.
And open weight too! So grateful for this.
Thank you for the link! Their playground in Mistral does not have a microphone. it just uploads files, which does not demonstrate the speed and accuracy, but the link you shared does.
I tried speaking in 2 languages at once, and it picked it up correctly. Truly impressive for real-time.
> Truly impressive for real-time.
Impressive indeed. Works way better than the speech recognition I first got demo'ed in... 1998? I remember you had to "click" on the mic everytime you wanted to speak and, well, not only the transcription was bad, it was so bad that it'd try to interpret the sound of the click as a word.
It was so bad I told several people not to invest in what was back then a national tech darling:
https://en.wikipedia.org/wiki/Lernout_%26_Hauspie
That turned out to be a massive fraud.
But ...
> I tried speaking in 2 languages at once, and it picked it up correctly.
I'm a native french speaker and I tried with a very simple sentence mixing french and english:
"Pour un pistolet je prefere un red dot mais pour une carabine je prefere un ACOG" (aka "For a pistol I prefer a red dot but for a carbine I prefer an ACOG")
And instead I got this:
"Je prépare un redote, mais pour une carabine, je préfère un ACOG."
"Je prépare un redote ..." doesn't mean anything and it's not at all what I said.
I like it, it's impressive, but literally the first sentence I tried it got the first half entirely wrong.
According to the announcement blog Le Chat is powered by the new model as well: https://chat.mistral.ai/chat
Doesn't seem to work for me - tried in both Firefox and Chromium and I can see the waveform when I talk but the transcription just shows "Awaiting audio input".
Try disabling CSP for the page
Same here. In Chromium I don't even see the waveform.
I had to turn off ad-block to get it to work.
404 on https://mistralai-voxtral-mini-realtime.hf.space/gradio_api/... for me (which shows up in the UI as a little red error in the top right).
It can transcribe Eminem's Rap God fast sequence, really, really impressive.
That's almost certainly in the training data, to be fair.
what a great test hahah
Wow, that’s weird. I tried Bengali, but the text transcribed into Hindi!I know there are some similar words in these languages, but I used pure Bengali that is not similar to Hindi.
Well, on the linked page, it mentions "strong transcription performance in 13 languages, including [...] Hindi" but with no mention of Bengali. It probably doesn't know a lick of Bengali, and is just trying to snap your words into the closest language it does know.
it must have some exposure to bengali— just not enough for them to advertise it. otherwise it would have a damn hard time.
This model was able to transcribe Bad Bunny lyrics over the sound of the background music, played casually from my speakers. Impressive, to me.
It's really nice although I've got a sentence in French when I was speaking Italian but I corrected myself in the middle of a word.
But I'm definitely going to keep an eye on this for local-only TTS for Home Assistant.
I’ve been using AquaVoice for real-time transcription for a while now, and it has become a core part of my workflow. It gets everything, jargon, capitalization, everything. Now I’m looking forward to doing that with 100% local inference!
Doesn’t seem to work in Safari on iOS 26.2, iPhone 17 Pro, just about anything extra disabled.
Here European Multilingual-Intelligence truly shines!
Not terrible. It missed or mixed up a lot of words when I was speaking quickly (and not enunciating very well), but it does well with normal-paced speech.
Yeah it messed up a bit for me too when I didn't enunciate well. If I speak clearly it seems to work very well even with background noise. Remember Dragon Naturally Speaking? Imagine having this back then!
is this demo running fully in the browser?
No, it's server-side.
Model is around 7.5 GB - once they get above 4 GB running them in a browser gets quite difficult I believe.
In English it is pretty good. But talk to it in Polish, and suddenly it thinks you speak Russian? Ukranian? Belarus? I would understand if an American company launched this, but for a company being so proud about their European roots, I think it should have better support for major European languages.
I tried English + Polish:
> All right, I'm not really sure if transcribing this makes a lot of sense. Maybe not. A цьому nie mówisz po polsku. A цьому nie mówisz po polsku, nie po ukrańsku.
They don't claim to support Polish, but they do support Russian.
> The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch. With a 4B parameter footprint, it runs efficiently on edge devices, ensuring privacy and security for sensitive deployments.
I wonder how much having languages with the same roots (e.g. the romance languages in the list above or multiple Slavic languages) affects the parameter count and the training set. Do you need more training data to differentiate between multiple similar languages? How would swapping, for example, Hindi (fairly distinct from the other 12 supported languages) for Ukrainian and Polish (both share some roots with Russian) affect the parameter count?
Nobody ever supports Polish. It's the worst. They'll support like, ̵Swahili, but not Polish.
200 million people speak Swahili.
39 million people speak Polish, and most of those also speak English or another more common language.
You could say the same about Dutch to be fair. 90-95% speak English - I bet that's way higher than in Poland.
Swahili is subcontinental lingua franca spoken by 200M people and growing quickly. Polish is spoken by a shrinking population in one country where English is understood anyways.
> The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch.
Try sticking to the supported languages
Yeah, it's too bad. Apparently it only performs well in certain languages: "The model is natively multilingual, achieving strong transcription performance in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch"
It did great English and Spanish, it didn't switch to Portuguese, french nor German, maybe struggle with my accent.
Try to warn it you are going to switch language to Portugese. Worked for me.
That's a mix of Polish and Ukrainian in the transcript. Now, if I try speaking Ukrainian, I'm getting transcript in Russian every time. That's upsetting.
Oh no! The model won't translate to an unsupported language, and incorrectly reverts to one that it was explicitly trained on.
The base likely was pretrained on days that included Polish and Ukrainian. You shouldn't be surprised to learn it doesn't perform great on languages it wasn't trained on, or perhaps had the highest share of training data.
Tell it you are going to speak Polish now. It helps.
TBH ChatGPT does the same, when I mix Polish and English. Generally getting some cyrillic characters and it gets super confused.
I'm not sure why but their multilingual performance in general has usually been below average. For a French company, their models are not even close to being best in French, even outdone by the likes of Qwen. I don't think they're focusing on anything but English, the rest is just marketing.
Incroyable! Competitive (if not better) than deepgram nova-3, and much better than assembly and elevenlabs in basically all cases on our internal streaming benchmarking.
The dataset is ~100 8kHz call recordings with gnarly UK accents (which I consider to be the final boss of english language ASR). It seems like it's SOTA.
Where it does fall down seems to be the latency distribution but I'm testing against the API. Running it locally will no doubt improve that?
> At approximately 4% word error rate on FLEURS and $0.003/min
Amazons transcription service is $0.024 per minute, pretty big difference https://aws.amazon.com/transcribe/pricing/
Is it 0.003 per minute of audio uploaded, or "compute minute"?
For example fal.ai has a Whisper API endpoint priced at "$0.00125 per compute second" which (at 10-25x realtime) is EXTREMELY cheaper than all the competitors.
I think the point is having it for real-time; this is for conversations rather than transcribing audio files.
That quote was for the non-realtime model.
Do we know if this is better than Nvidia Parakeet V3? That has been my go-to model locally and it's hard to imagine there's something even better.
I've been using nemotron ASR with my own ported inference, and happy about it:
https://huggingface.co/nvidia/nemotron-speech-streaming-en-0...
https://github.com/m1el/nemotron-asr.cpp https://huggingface.co/m1el/nemotron-speech-streaming-0.6B-g...
I'm so amazed to find out just how close we are to the start trek voice computer.
I used to use Dragon Dictation to draft my first novel, had to learn a 'language' to tell the rudimentary engine how to recognize my speech.
And then I discovered [1] and have been using it for some basic speech recognition, amazed at what a local model can do.
But it can't transcribe any text until I finish recording a file, and then it starts work, so very slow batches in terms of feedback latency cycles.
And now you've posted this cool solution which streams audio chunks to a model in infinite small pieces, amazing, just amazing.
Now if only I can figure out how to contribute to Handy or similar to do that Speech To Text in a streaming mode, STT locally will be a solved problem for me.
[1] https://github.com/cjpais/Handy
you should check out
https://github.com/pipecat-ai/nemotron-january-2026/
discovered through this twitter post:
https://x.com/kwindla/status/2008601717987045382
Happy to answer questions about this (or work with people on further optimizing the open source inference code here). NVIDIA has more inference tooling coming, but it's also fun to hack on the PyTorch/etc stuff they've released so far.
I've been using Parakeet V3 locally and totally ancedotaly this feels more accurate but slightly slower
I liked Parakeet v3 a lot until it started to drop whole sentences, willy-nilly.
Yeah, I think the multilingual improvements in V3 caused some kind of regression for English - I've noticed large blocks occasionally dropped as well, so reverted to v2 for my usage. Specifically nvidia/parakeet-tdt-0.6b-v2 vs nvidia/parakeet-tdt-0.6b-v3
Parakeet is really good imo too, and it's just 0.6B so it can actually run on edge devices. 4B is massive, I don't see Voxtral running realtime on an Orin or fitting on a Hailo. An Orin Nano probably can't even load it at BF16.
Came here to ask the same question!
I noticed that this model is multilingual and understands 14 languages. For many use cases, we probably only need a single language, and the extra 13 are simply adding extra latency. I believe there will be a trend in the coming years of trimming the fat off of these jack of all trades models.
https://aclanthology.org/2025.findings-acl.87/
I don't know. What about words inherited from other languages? I think a cross-language model could improve lots of things.
For example, "here it is, voila!" "turn left on el camino real"
I think this model proves it's very efficient and accurate.
But it could potentially be even more efficient if it was single-language.
STT services that have been around for longer, like Azure, Google and Amazon, generally require you to request a specific language, and their quality is a lot higher than models that advertise themselves as LLMs (even though I believe the clouds are also using the same types of models now).
It doesn't make sense to have a language-restricted transcription model because of code switching. People aren't machines, we don't stick to our native languages without failure. Even monolingual people move in and out of their native language when using "borrowed" words/phrases. A single-language model will often fail to deal with that.
yeah, one example I run into is getting my perplexity phone assistant to play a song in spanish. I cannot for the life of me get a model to translate: "Play señorita a mi me gusta su style on spotify" correctly
The hilarious part of this comment is all the comments around it complaining about not supporting enough languages
uhhh i cast doubt on multi-language support as affecting latency. model size, maybe, but what is the mechanism for making latency worse? i think of model latency as O(log(model size))… but i am open to being wrong / that being a not-good mental model / educated guess.
Even model size, it’s modest. There is a lot of machinery that is going to be common for all languages. You don’t multiply model size by 2 when you double the number of supported languages.
If encoding more learned languages and grammars and dictionaries makes the model size bigger, it will also increase latency. Try running a 1B model locally and then try to run a 500B model on the same hardware. You'll notice that latency has rather a lot to do with model size.
model size directly affects latency
Imagine if ChatGPT started like this and thought they should trim coding abilities from their language model because most people don't code.
They've already done the inverse and trimmed non-coding abilities from their language model: https://openai.com/index/introducing-gpt-5-2-codex/. There's already precedent for creating domain-specific models.
I think it's nice to have specialized models for specific tasks that don't try to be generalists. Voxtral Transcript 2 is already extremely impressive, so imagine how much better it could be if it specialized in specific languages rather than cramming 14 languages into one model.
That said, generalist models definitely have their uses. I do want multilingual transcribing models to exist, I just also think that monolingual models could potentially achieve even better results for that specific language.
Native diarization, this looks exciting. edit: or not, no diarization in real-time.
https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-26...
~9GB model.
The diarization is on Voxtral Mini Transcribe V2, not Voxtral Mini 4B.
Ahh, yeah, and it's explicitly not working for realtime streams. Good catch!
Do you have experience with that model for diarization? Does it feel accurate, and what's its realtime factor on a typical GPU? Diarization has been the biggest thorn in my side for a long time..
You can test it yourself for free on https://console.mistral.ai/build/audio/speech-to-text I tried it on an english-speaking podcast episode, and apart from identying one host as two different speakers (but only once for a few sentences at the start), the rest was flawless from what I could see
Amazing. Thank you.
> Do you have experience with that model
No, I just heard about it this morning.
https://www.tavus.io/post/sparrow-1-human-level-conversation...
how does it compare to sparrow-1?
Played with the demo a bit. It's really good at English, and detects language change on the fly. Impressive.
But whatever I tried, it could not recognise my Ukrainian and would default to Russian in absolutely ridiculous transcription. Other STT models recognise Ukrainian consistently, so I assume there is a lot of Russian in training material, and zero Ukrainian. Made me really sad.
Thats just the result of the model only supporting russian (and 12 other languages) and not urkainian. It maps to the closest words from training data.
It’s nice, but the previous version wasn’t actually that great compared to Parakeet for example.
We need better independent comparison to see how it performs against the latest Qwen3-ASR, and so on.
I can no longer take at face value the cherry picked comparisons of the companies showing off their new models.
For now, NVIDIA Parakeet v3 is the best for my use case, and runs very fast on my laptop or my phone.
There is https://huggingface.co/spaces/hf-audio/open_asr_leaderboard but it hasn't been updated for half a year.
I like Parakeet as well and use it via Handy on Mac. What app are you using on your phone?
Spokenly has it on Mac and iOS, in both cases for free when using parakeet
Is there an open source Android keyboard that would support it? Everything I find is based on Whisper, which is from 2022. Ages ago given how fast AI is evolving.
I wish I had a Google Keyboard that could easily run on Whisper Medium. This is already great. But unfortunately would be too much inference cost, incredibly slow. The problem with Whisper is not the inference quality: medium and large are incredible. Is that the base model is not enough, and the only one with fast inference in mobile devices.
There's no comparison to Whisper Large v3 or other Whisper models..
Is it better? Worse? Why do they only compare to gpt4o mini transcribe?
WER is slightly misleading, but Whisper Large v3 WER is classically around 10%, I think, and 12% with Turbo.
The thing that makes it particularly misleading is that models that do transcription to lowercase and then use inverse text normalization to restore structure and grammar end up making a very different class of mistakes than Whisper, which goes directly to final form text including punctuation and quotes and tone.
But nonetheless, they're claiming such a lower error rate than Whisper that it's almost not in the same bucket.
On the topic of things being misleading, GPT-4o transcriber is a very _different_ transcriber to Whisper. I would say not better or worse, despite characterizations such. So it is a little difficult to compare on just the numbers.
There's a reason that quite a lot of good transcribers still use V2, not V3.
Different how?
Gpt4o mini transcribe is better and actually realtime. Whisper is trained to encode the entire audio (or at least 30s chunks) and then decode it.
So "gpt4o mini transcribe" is not just whisper v3 under the hood? Btw it's $0.006 / minute
For Whisper API online (with v3 large) I've found "$0.00125 per compute second" which is the cheapest absolute I've ever found.
Deepinfra offers Whisper V3 at 0.00045$ / minute of transcribed audio.
>So it's not just whisper v3 under the hood?
Why it should be Whisper v3? They even released an open model: https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-26...
The linked article claims the average word error rate for Voxtral mini v2 is lower than GPT-4o mini transcribe
Gpt4o mini transcribe is better than whisper, the context is the parent comment.
3 hours for a single request sounds nice to me. Although the graph suggests that it’s not going to perform as good as openai model I have been using, it is open source and surely I will give it a try.
I really wish those offering speech-to-text models provided transcription benchmarks specific to particular fields of endeavor. I imagine performance would vary wildly when using jargon peculiar to software development, medicine, physics, and law, as compared to everyday speech. Considering that "enterprise" use is often specialized or sub-specialized, it seems like they're leaving money on Dragon's table by not catering to any of those needs.
Looks like this model doesn't do realtime diarization, what model should I use if I want that? So far I've only seen paid models do diarization well. I heard about Nvidia NeMo but haven't tried that or even where to try it out.
Not sure if its "realtime" but the recently released VibeVoice-ASR from Microsoft does do diarization. https://huggingface.co/microsoft/VibeVoice-ASR
things I hate:
"Click me to try now!" banners that lead to a warning screen that says "Oh, only paying members, whoops!"
So, you don't mean 'try this out', you mean 'buy this product'.
Let's not act like it's a free sampler.
I can't comment on the model : i'm not giving them money.
You can try it on HF: https://huggingface.co/spaces/mistralai/Voxtral-Mini-Realtim...
I'm impressed.
Is it me or error rate of 3% is really high?
If you transcribe a minute of conversation, you'll have like 5 words transcribed wrongly. In an hour podcast, that is 300 wrongly transcribed words.
The error rate for human transcription can be as high as 5%.
Oh wow, I thought humans are like 0.1% error rate, if they are native speakers and aware of the subject being discussed.
I was skepitcal upon hearing the figure but various sources do indeed back it up and [0] is a pretty interesting paper (old but still relevant human transcibers haven't changed in accuracy).
[0] https://www.microsoft.com/en-us/research/wp-content/uploads/...
I think it's actually hard to verify how correct a transcription is, at scale. Curious where those error rate numbers come from, because they should test it on people actually doing their job.
It can depend a lot on different factors like:
- familiarity with the accent and/or speaker;
- speed and style/cadence of the speech;
- any other audio that is happening that can muffle or distort the audio;
- etc.
It can also take multiple passes to get a decent transcription.
Most of these errors will not be meaningful. Real speech is full of ambiguities. 3% is low
What's the cheapest device specs that this could realistically run on?
I haven't quite figured out if the open weights they released on huggingface amount to being able to run the (realtime) model locally - i hope so though! For the larger model with diarization I don't think they open sourced anything.
The HF page suggests yes, with vllm.
> We've worked hand-in-hand with the vLLM team to have production-grade support for Voxtral Mini 4B Realtime 2602 with vLLM. Special thanks goes out to Joshua Deng, Yu Luo, Chen Zhang, Nick Hill, Nicolò Lucchesi, Roger Wang, and Cyrus Leung for the amazing work and help on building a production-ready audio streaming and realtime system in vLLM.
https://huggingface.co/mistralai/Voxtral-Mini-4B-Realtime-26...
https://docs.vllm.ai/en/latest/serving/openai_compatible_ser...
Italian represents, I believe, the most phonetically advanced human language. It has the right compromise among information density, understandability, and ability to speech much faster to compensate the redundancy. It's like if it had error correction built-in. Note that it's not just that it has the lower error rate, but is also underrepresented in most datasets.
I love seeing people from other countries share their own folk tales about what makes their countries special and unique. I've seen it up close in my country and I always cringed when I heard my fellow countrymen came up with these stories. In my adulthood I'm reassured that it happens everywhere and I find it endearing.
On the information density of languages: it is true that some languages have a more information dense textual representation. But all spoken languages convey about the same information in the same time. Which is not all that surprising, it just means that human brains have an optimal range at which they process information.
Further reading: Coupé, Christophe, et al. "Different Languages, Similar Encoding Efficiency: Comparable Information Rates across the Human Communicative Niche." Science Advances. https://doi.org/10.1126/sciadv.aaw2594
Different representations at the same bitrate may have features that make one a lot more resilient to errors. This thing about Italian, you fill find in any benchmark of vastly different AI transcribing models. You can find similar results also on the way LLMs mostly trained on English generalize usually very well with Italian. All this despite Italian accounting for marginal percentage of the training set. How do you explain that? I always cringe when people refute evidence.
> All this despite Italian accounting for marginal percentage of the training set.
Evidence?
Where is this evidence you’ve cited for your claims?
This is largely due to the fact that modern Italian is a systematised language that emerged from a literary movement (whose most prominent representative is Alessandro Manzoni) to establish a uniform language for the Italian people. At the time of Italian unification in 1861, only about 2.5% of the population could speak this language.
The language itself was not invented for the purpose: it was the language spoken in Florence, than adopted by the literary movement and than selected as the national language.
It seems like the best tradeoff between information density and understandability actually comes from the deep latin roots of the language
in the end (our) italian language wasn’t optimized by engineers, it was refactored by poets
I was honestly surprised to find it in the first place, because I assumed English to be at first place given the simpler grammar and the huge dataset available.
I agree with your belief, other languages have either lower density (e.g. German) or lower understandability (e.g. English)
English has a ton of homophones, way more sounds that differ slightly (long/short vowels), and major pronunciation differences across major "official" languages (think Australia/US/Canada/UK).
Italian has one official italian (two, if you count IT_ch, but difference is minor), doesn't pay much attention to stress and vowel length, and only has a few "confusable" sounds (gl/l, gn/n, double consonants, stuff you get wrong in primary school). Italian dialects would be a disaster tho :)
The only knowledge I have about how difficult Italian is comes from Inglourious Basterds.
> the most phonetically advanced human language
That's interesting. As a linguist, I have to say that Haskell is the most computationally advanced programming language, having the best balance of clear syntax and expressiveness. I am qualified to say this because I once used Haskell to make a web site, and I also tried C++ but I kept on getting errors.
/s obviously.
Tldr: computer scientists feel unjustifiably entitled to make scientific-sounding but meaningless pronouncements on topics outside their field of expertise.
At least some relatively well-known research finds that all languages have similar information density in terms of bits/second (~39 bits/second based on a quick search). Languages do it with different amounts of phonetic sound / syllables / words per bit and per second, but the bps comes out the same.
I don't know how widely accepted that conclusion is, what exceptions there may be, etc.
What's the best way to train this further on a specific dialect or accent or even terminology?
This looks great, but it's not clear to me how to use it for a practical task. I need to transcribe about 10 years worth of monthly meetings. These are government hearings with a variety of speakers. All the videos are on YouTube. What's the most practical and cost-effective way to get reasonably accurate transcripts?
If you use something like youtube-dlp you can download the audio from the meetings, and you could try things out in mistrals ai studio.
You could use their api (they have this snippet):
```curl -X POST "https://api.mistral.ai/v1/audio/transcriptions" \ -H "Authorization: Bearer $MISTRAL_API_KEY" \ -F model="voxtral-mini-latest" \ -F file=@"your-file.m4a" \ -F diarize=true \ -F timestamp_granularities="segment"```
In the api it took 18s to do a 20m audio file I had lying around where someone is reviewing a product.
There will, I'm sure, be ways of running this locally up and available soon (if they aren't in huggingface right now) but the API is $0.003/min. If it's something like 120 meetings (10 years of monthly ones) then it's roughly $20 if the meetings are 1hr each. Depending on whether they're 1 or 10 hours (or if they're weekly or monthly but 10 parallel sessions or something) then this might be a price you're willing to pay if you get the results back in an afternoon.
edit - their realtime model can be run with vllm, the batch model is not open
- get an API key for this service
- make sure you have a list of all these YouTube meeting URLs somewhere
- ask your preferred coding assistant to write you up a script that downloads the audio for these videos with yt-dlp & calls Mixtrals' API
- ????
- profit
If they are on Youtube, try Gemini 3 Flash first. Use AI studio, it lets you insert YouTube videos into context.
As a rule of thumb for software that I use regularly, it is very useful to consider the costs over a 10-year period in order to compare it with software that I purchase for lifetime to install at home. So that means 1,798.80 $ for the Pro version.
What estimates do others use?
Wired advertises this as "Ultra-Fast Translation"[^1]. A bit weird coming from a tech magazine. I hope it's just a "typo".
[^1]: https://www.wired.com/story/mistral-voxtral-real-time-ai-tra...
It might be capable of translation; OpenAI Whisper was a transcription model that could do it.
One week ago I was on the hunt for an open source model that can do diatization and I had to literally give up because I could not find any easy to use setup.
I don't know if that will change, but right now only the Voxtral Mini Transcribe V2 supports diarization and it's not open-weight. The Voxtral Realtime model doesn't support diarization, but is open-weight.
WhisperX ?
I'm guessing I won't be able to finetune this until they come out with a HF tranformers model, right?
Impressive results, tested on crappy audio files (in french and english)...
does anyone know if there's any desktop tools I can use this transcription model with? e.g. something where like Wisper Flow/WillowVoice but with custom model selection
There is Handy, an open source project meant to be a desktop tool, but I haven’t installed it yet to see how you pick your model.
Handy – Free open source speech-to-text app https://github.com/cjpais/Handy
I added it to my bot agent,let’s see how it performs
Nice. Can this be ran on a mobile device?
Disappointing how this lacks a clear reference implementation, if not mixed at almost yet unreleased VLLM (nightly version) stuff. I'm ok with Open Weights being a form of OSS in the case of models, because frankly I don't believe that, for large LLMs, it is feasible to release the training data, all the orchestration stuff, and so forth. But it can't be: here are the weights, we partnered with VLLM for inference. Come on. Open Weights must mean that you put me in a situation to write an implementation easily for any hardware.
p.s. even the demo uses a remote server via websocket.
Any chance Voxtral Mini Transcribe 2 will ever be an open model?
Can it translate in real time?
Also curious about this. Just need real time German to English. What does this?
Do you know anything better for Polish language, low quality audio than Whisper large-v3 through WhisperX?
This combo has almost unbeatable accuracy and it rejects noises in the background really well. It can even reject people talking in the background.
The only better thing I've seen is Ursa model from Speechmatics. Not open weights unfortunately.
I'm on voxtral-mini-latest and that's why I started seeing 500s today lol
Pseudo related -- am I the only one uncomfortable using my voice with AI for the concern that once it is in the training model it is forever reproducible? As a non-public person it seems like a risk vector (albeit small),
It's a real issue, but why do you only see it in ai? It's true for any case where you're speaking into a microphone
Depending on the permissions granted to apps on your mobile device, it can even be passively exfiltrated without you ever noticing - and that's ignoring the video clips people take and put online. Like your grandma uploading to Facebook a short moment from a Christmas meet or similar
There have already been successful scams - eg calls from "relatives" (AI) calling family members needing money urgently and convincing them to send the money...