If these are the "gpt-4o-mini-tts" models, and if the pricing estimate of "$0.015 per minute" of audio is correct, then these prices 85% cheaper than those of ElevenLabs.
With ElevenLabs, if I choose their most cost-effectuve "Business" plan for $1100 per month (with annual billing of $13,200, a savings of 17% over monthly billing), then I get 11,000 minutes TTS, and each minute is billed at 10 cents.
With OpenAI, I could get 11,000 minutes of TTS for $165.
It's way cheaper - everyone is, elevenlabs is very expensive. Nobody matches their quality though. Especially if you want something that doesn't sound like a voice assistant/audiobook/podcast/news anchor/tv announcer.
This openai offering is very interesting, it offers valuable features elevenlabs doesn't in emotional control. It also hallucinates though which would need to be fixed for it to be very useful.
It's cheap because everything OpenAI does is subsidized by investors' money.
Until that stupid money flows all good!
Then either they'll go the way of WeWork, or enshittification will happen to make it possible for them to make the books work.
I don't see any other option.
Unless Softbank decides it has some 150 Billion to squander on buying them off.
There's a lot of head-in-the-sand behavior going on around OpenAI fundamentals and I don't understand exactly why it's not more in the open yet.
Elevenlabs is an ecosystem play. They have hundreds of different voices, legally licensed from real people who chose to upload their voice. It is a marketplace of voices.
None of the other major players is trying to do that, not sure why.
ElevenLabs is the only one offering speech to speech generation where the intonation, prosody, and timing is kept intact. This allows for one expressive voice actor to slip into many other voices.
I hope they find a more unique product offering that takes hold. Everybody thinks of them as text-to-speech but I use ElevenLabs exclusively for speech-to-speech for vtubing as my AI character. They're kind of the only game in town for doing super high quality speech-to-speech (unless someone here has an alternative which I'd LOVE to know about). I've tried https://github.com/w-okada/voice-changer which is great because it's real-time but the quality is enough of a step down that actual words I'm saying become unclear and difficult to understand. Also with that I am tied to using my RTX 3090 desktop vs ElevenLabs which I can do in the cloud from my laptop anywhere.
I'm pretty much dependent on ElevenLabs to do my vtubing at this point but I can't imagine speech-to-speech has wide adoption so I don't know if they'll even keep it around.
Yes ElevenLabs is orders of magnitude more expensive than everyone else. Very clever from a business perspective, I think. They are (were?) the best so know that people will pay a premium for that.
Yeah the way I see it this is where we find the value of customization. We are already seeing its use by YouTube video essay creators who turn their own voice into models. I want to see corporate executives get on board so that we can finally ditch the god awful phone quality in earnings calls.
yes, I think you are right. When I did the math on 11labs million chars I got the same numbers (Pro plan).
I'm super happy about this, since I took a bet that exactly this would happen. I've just been building a consumer TTS app that could only work with significant cheaper TTS prices per million character (or self-hosted models)
Hey, I'm Jeff and I was PM for these models at OpenAI. Today we launched three new state-of-the-art audio models. Two speech-to-text models—outperforming Whisper. A new TTS model—you can instruct it how to speak (try it on openai.fm!). And our Agents SDK now supports audio, making it easy to turn text agents into voice agents. We think you'll really like these models. Let me know if you have any questions here!
Hi Jeff. This is awesome. Any plans to add word timestamps to the new speech-to-text models, though?
> Other parameters, such as timestamp_granularities, require verbose_json output and are therefore only available when using whisper-1.
Word timestamps are insanely useful for large calls with interruptions (e.g. multi-party debate/Twitter spaces), allowing transcript lines to be further split post-transcription on semantic boundaries rather than crude VAD-detected silence. Without timestamps it’s near-impossible to make intelligible two paragraphs from Speaker 1 and Speaker 2 with both interrupting each other without aggressively partitioning source audio pre-transcription—which severely degrades transcript quality, increases hallucination frequency and still doesn’t get the same quality as word timestamps. :)
Accurate word timestamps seems an overhead and required a post processing like forced alignment (speech technique that can automatically align audio files with transcripts)
Had a recent dive into a forced alignment, and discovered that most of new models dont operate on word boundaries, phoneme, etc but rather chunk audio with overlap and do word, context matching. Older HHM-style models have shorter strides (10ms vs 20ms).
Tried to search into Kaldi/Sherpa ecosystem, and found most info leads to nowhere or very small and inaccurate models.
Having read the docs - used chat gpt to summarize them - there is no mention of speaker diarization for these models.
This is a _very_ low hanging fruit anyone with a couple of dgx h100 servers can solve in a month and is a real world problem that needs solving.
Right now _no_ tools on the market - paid or otherwise - can solve this with better than 60% accuracy. One killer feature for decision makers is the ability to chat with meetings to figure out who promised what, when and why. Without speaker diarization this only reliably works for remote meetings where you assume each audio stream is a separate person.
In short: please give us a diarization model. It's not that hard - I've done it one for a board of 5, with a 4090 over a weekend.
> This is a _very_ low hanging fruit anyone with a couple of dgx h100 servers can solve in a month and is a real world problem that needs solving.
I am not convinced it is a low hanging fruit, it's something that is super easy for humans but not trivial for machines, but you are right that it is being neglected by many. I work for speechmatics.com and we spent a significant amoutn of effort over the years on it. We now believe we have the world's best real-time speaker diarization system, you should give it a try.
Hi Jeff, thanks for these and congrats on the launch. Your docs mention supporting accents. I cannot get accents to work at all with the demo.
For instance erasing the entire instruction and replacing it with ‘speak with a strong Boston accent using eg sounds like hahhvahhd’ has no audible effect on the output.
As I’m sure you know 4o at launch was quite capable in this regard, and able to speak in a number of dialects and idiolects, although every month or two seems to bring more nerfs sadly.
A) can you guys explain how to get a US regional accent out of the instructions? On what you meant by accent if not that?
B) since you’re here I’d like to make a pitch that setting 4o for refusal to speak with an AAVE accent probably felt like a good idea to well intentioned white people working in safety. (We are stopping racism! AAVE isn’t funny!) However, the upshot is that my black kid can’t talk to an ai that sounds like him. Well, it can talk like he does if he’s code switching to hang out with your safety folks, but it considers how he talks with his peers as too dangerous to replicate.
This is a pernicious second order race and culture impact that I think is not where the company should be.
I expect this won’t get changed - chat is quite adamant that talking like millions of Americans do would be ‘harmful’ - but it’s one of those moments where I feel the worst parts of the culture wars coming back around to create the harm it purports to care about.
Anyway the 4o voice to voice team clearly allows the non mini model to talk like a Bostonian which makes me feel happy and represented; can the mini api version do this?
> e.g. the audio-preview model when given instruction to speak "What is the capital of Italy" would often speak "Rome". This model should be much better in that regard
"Much better" doesn't sound like it can't happen at all though.
1) Previous TTS models had problems with major problems accents. E.g. a Spanish sentence could drift from a Spain accent to Mexican to American all within one sentence. Has this been improved and/or is it still a WIP?
2) What is the latency?
3) Your STT API/Whisper had MAJOR problems with hallucinating things the user didn't say. Is this fixed?
4) Whisper and your audio models often auto corrected speech, e.g. if someone made a grammatical error. Or if someone is speaking Spanish and inserted an English word, it would change the word to the Spanish equivalent. Does this still happen?
1/ we've been working a lot on accents, so expect improvements with these models... though we're not done. Would be curious how you find them. And try giving specific detailed instructions + examples for the accents you want
2/ We're doing everything we can to make it fast. Very critical that it can stream audio meaningfully faster than realtime
3+4/ I wouldn't call hallucinations "solved", but it's been the central focus for these models. So I hope you find it much improved
Hi Jeff, are there any plans to support dual-channel audio recordings (e.g., Twilio phone call audio) for speech-to-text models? Currently, we have to either process each channel separately and lose conversational context, or merge channels and lose speaker identification.
Hey Jeff, this is awesome! I’m actually building a S2S application right now for a startup with the Realtime API and keen to know when these new voices/expressive prompting will be coming to it?
Also, any word on when there might be a way to move the prompting to the server side (of a full stack web app)? At the moment we have no way to protect our prompts from being inspected in the browser dev tools — even the initial instructions when the session is initiated on the server end up being spat back out to the browser client when the WebRTC connection is first made! It’s damaging to any viable business model.
Do you have plans to make it more realistic like kokoro-82M? I don't know, is it only me or anyone else, machine voice is irritating to me to listen for longer period of time.
How is the latency (Time To First Byte of audio, when streaming) and throughput (non-vibe characters input per second) compared to the existing 'tts-1' non-HD that's the same price? TTFB in particular is important and needs to be much better than 'tts-1'.
Hi Jeff, Thanks for updating the TTS endpoint! I was literally about to have to make a workaround with the chat completions endpoint with a hit and hope the transcription matches strategy... as it was the only way to get the updated voice models.
Curious.. is gpt-4o-mini-tts the equivilant of what is/was gpt-4o-mini-audio-preview for chat completions? Because in timing tests it takes around 2 seconds to return a short phrase which seems more equivilant to gpt-4o-audio-preview.. the later was much better for the hit and hope strat as it didn't ad lib!
Also I notice you can add accents to instructions and it does a reasonable job. But are there any plans to bring out localized voice models?
It's a slightly better model for TTS. With extra training focusing on reading the script exactly as written.
e.g. the audio-preview model when given instruction to speak "What is the capital of Italy" would often speak "Rome". This model should be much better in that regard
=
No plans to have localized voice models, but we do want to bring expand the menu of voices with voices that are best at different accents
Woohoo new voices! I’ve been using a mix of TTS models on a project I’ve been working on, and I consistently prefer the output of OpenAI to ElevenLabs (at least when things are working properly).
Which leads me to my main gripe with the OpenAI models — I find they break — produce empty / incorrect / noise outputs — on a few key use cases for my application (things like single-word inputs — especially compound words and capitalized words, words in parenthesis, etc.)
So I guess my question is might gpt-4o-mini-tts provide more “reliable” output than tts-1-hd?
Yes, from our terms: "Don’t build tools that may be inappropriate for minors, including: Sexually explicit or suggestive content. This does not include content created for scientific or educational purposes." https://openai.com/policies/usage-policies/
Do you know when we can expect an update on the realtime API? It’s still in beta and there are many issues (e.g voice randomly cutting off, VAD issues, especially with mulaw etc…) which makes it impossible to use in production, but there’s not much communication from OpenAI. It’s difficult to know what to bet on. Pushing for stt->llm->tts makes you wonder if we should carry on building with the realtime API.
we're working hard on it at the moment and hope we'll have a snapshot ready in the next month or so
we've debugged the cutoff issues and have fixes for them internally but we need a snapshot that's better across the board, not just cutoffs (working on it!)
we're all in on S2S models both for API and ChatGPT, so there will be lots more coming to Realtime this year
For today: the new noise cancellation and semantic voice activity detector are available in Realtime. And ofc you can use gpt-4o-transribe for user transcripts there
Hi Jeff, I have an app that already supports the Whisper API, so I added the GPT4o models as options. I noticed that the GPT4o models don't support prompting, and as a result my app had a higher error rate in practice when using GPT4o compared to Whisper. Is prompting on the roadmap?
Hey Jeff, maybe you could improve the TTS that is currently in the OpenAI web and phone apps. When I set it to read numbers in Romanian it slurs digits. This also happens sometimes with regular words as well. I hope you find resources for other languages than English.
thanks for flagging ... number fidelity (especially on languages that are unfortunately less represented in training data) is still something we're working to improve
How about more sample code for the streaming transcription api? I gave o1pro the docs for both the real-time endpoint and the stt API but we couldn't get it working (from Java, but any language would help).
Please release a stable realtime speech to speech model. The current version constantly thinks it’s a young teen heading to college and sad but then suddenly so excited about it
Hey Jeff, thanks for your work! Quick question for you, are you guys using Azure Speech Services or have these TTS models been trained by OpenAI from scratch?
After toying around with the TTS model it seems incredibly nondeterministic. Running the same input with the same parameters can have widely different results, some really good, others downright bad. The tone, intonation and character all vary widely. While some of the outputs are great, this inconsistency makes it a really tough sell. Imagine if Siri responded to you with a different voice every time, as an example. Is this something you're looking to address somewhere down the line or do you consider that working as intended?
Whisper's major problem was hallucinations, how are the new models doing there? The performance of ChatGPT advanced voice in recognizing speech is, frankly, terrible. Are these models better than what's used there?
How did you make whisper better? I used whisper large to transcribe 30 podcast episodes and it did an amazing job. The times it made mistakes were understandable like confusing “Macs” and “Max”, slurred speech or people just saying things in a weird way. I was able to correct these mistakes because I understood the context of what was being talked about.
Another thing I noticed is whisper did a better job of transcribing when I removed a lot of the silences in the audio.
Both the text-to-speech and the speech-to-text models launched here suffer from reliability issues due to combining instructions and data in the same stream of tokens.
Thanks for the write up. I've been writing assembly lately, so as soon as I read your comment, I thought "hmm reminds me of section .text and section .data".
Large text-to-speech and speech-to-text models have been greatly improving recently.
But I wish there were an offline, on-device, multilingual text-to-speech solution with good voices for a standard PC — one that doesn't require a GPU, tons of RAM, or max out the CPU.
In my research, I didn't find anything that fits the bill. People often mention Tortoise TTS, but I think it garbles words too often. The only plug-in solution for desktop apps I know of is the commercial and rather pricey Acapela SDK.
I hope someone can shrink those new neural network–based models to run efficiently on a typical computer. Ideally, it should run at under 50% CPU load on an average Windows laptop that’s several years old, and start speaking almost immediately (less than 400ms delay).
The same goes for speech-to-text. Whisper.cpp is fine, but last time I looked, it wasn't able to transcribe audio at real-time speed on a standard laptop.
I'd pay for something like this as long as it's less expensive than Acapela.
The sample sounds impressive, but based on their claim -- 'Streaming inference is faster than playback even on an A100 40GB for the 3 billion parameter model' -- I don't think this could run on a standard laptop.
Thanks! But I get the impression that with Kokoro, a strong CPU still requires about two seconds to generate one sentence, which is too much of a delay for a TTS voice in an AAC app.
I'd rather accept a little compromise regarding the voice and intonation quality, as long as the TTS system doesn't frequently garble words. The AAC app is used on tablet PCs running from battery, so the lower the CPU usage and energy draw, the better.
I use Piper for one of my apps. It runs on CPU and doesn't require a GPU. It will run well on a raspberry pi. I found a couple of permissively licensed voices that could handle technical terms without garbling them.
However, it is unmaintained and the Apple Silicon build is broken.
My app also uses whisper.cpp. It runs in real time on Apple Sillicon or on modern fast CPUs like AMD's gaming CPUs.
I had already suspected that I hadn't found all the possibilities regarding Tortoise TTS, Coqui, Piper, etc. It is sometimes difficult to determine how good a TTS framework really is.
Do you possibly have links to the voices you found?
Is there way to get "speech marks" alongside the generated audio?
FYI, Speech marks provide millisecond timestamp for each word in a generated audio file/stream (and a start/end index into your original source string), as a stream of JSONL objects, like this:
AWS uses these speech marks (with variants for "sentence", "word", "viseme", or "ssml") in their Polly TTS service...
The sentence or word marks are useful for highlighting text as the TTS reads aloud, while the "viseme" marks are useful for doing lip-sync on a facial model.
This is astonishing. I can type anything I want into the "vibe" box and it does it for the given text. Accents, attitudes, personality types... I'm amazed.
The level of intelligent "prosody" here -- the rhythm and intonation, the pauses and personality -- I wasn't expecting anything like this so soon. This is truly remarkable. It understands both the text and the prompt for how the speaker should sound.
Like, we're getting much closer to the point where nobody except celebrities are going to record audiobooks. Everyone's just going to pick whatever voice they're in the mood for.
Some fun ones I just came up with:
> Imposing villain with an upper class British accent, speaking threateningly and with menace.
> Helpful customer support assistant with a Southern drawl who's very enthusiastic.
> Woman with a Boston accent who talks incredibly slowly and sounds like she's about to fall asleep at any minute.
I don't see how a strike will do anything but accelerate the professions inevitable demise. Can anyone explain how this could ever end in favor of the human laborers striking?
> Everyone's just going to pick whatever voice they're in the mood for.
I can't say I've ever had this impulse. Also, to point out the obvious, there's little reason to pay for an audiobook if there's no human reading it. Especially if you already bought the physical text.
Didn’t look closely, but is there a way to clone a voice from a few seconds of recording and then feed the sample to generate the text in the same voice?
I am always listening to audio books but they are no good anymore after playing with this for 2 minutes.
I am never really in the mood for a different voice. I am going to dial in the voice I want and only going to want to listen with that voice.
This is so awesome. So many audio books have been ruined by the voice actor for me. What sticks out in my head is The Book of Why by Judea Pearl read by Mel Foster. Brutal.
So many books I want as audio books too that no one would bother to record.
The ElevenReader app from ElevenLabs has been able to do that for a while now and they’ve licensed some celebrity voices like Burt Reynolds. You can use the browser share function to send it a webpage to read or upload a PDF or epub of a book.
It’s far from perfect though. I’m listening to Shattered Sword (about the battle of midway) which has lots of academic style citations so every other sentence or paragraph ends with it spelling out the citation number like “end of sentence dot one zero”, it’ll often mangle numbers like “1,000 pound bomb” becomes “one zero zero zero pound bomb”, and it tries way too hard to expand abbreviations so “Operation AL” becomes “Operation Alabama” when it’s really short for Aleutian Islands.
One very important quote from the official announcement:
> For the first time, developers can “instruct” the model not just on what to say but how to say it—enabling more customized experiences for use cases ranging from customer service to creative storytelling.
The instructions are the "vibes" in this UI. But the announcement is wrong with the "for the first time" part: it was possible to steer the base GPT-4o model to create voices in a certain style using system prompt engineering (blogged about here: https://minimaxir.com/2024/10/speech-prompt-engineering/ ) out of concern that it could be used as a replacement for voice acting, however it was too expensive and adherence isn't great.
The schema of the vibes here implies that this new model is more receptive to nuance, which changes the calculus. The test cases from my post behave as expected, and the cost of gpt-4o-mini-tts audio output is $0.015 / minute (https://platform.openai.com/docs/pricing ), which is about 1/20th of the cost of my initial experments and is now feasible to use to potentially replace common voice applications. This has implications, and I'll be testing more around more nuanced prompt engineering.
I gave it (part of) the classic Navy Seal copypasta.
Interestingly, the safety controls ("I cannot assist with that request") is sort of dependent on the vibe instruction. NYC cabbie has no problem with it (and it's really, really funny, great job openAI), but anything peaceful, positive, etc. will deny the request.
Readit News
https://platform.openai.com/docs/pricing
If these are the "gpt-4o-mini-tts" models, and if the pricing estimate of "$0.015 per minute" of audio is correct, then these prices 85% cheaper than those of ElevenLabs.
https://elevenlabs.io/pricing
With ElevenLabs, if I choose their most cost-effectuve "Business" plan for $1100 per month (with annual billing of $13,200, a savings of 17% over monthly billing), then I get 11,000 minutes TTS, and each minute is billed at 10 cents.
With OpenAI, I could get 11,000 minutes of TTS for $165.
Somebody check my math... Is this right?
This openai offering is very interesting, it offers valuable features elevenlabs doesn't in emotional control. It also hallucinates though which would need to be fixed for it to be very useful.
None of the other major players is trying to do that, not sure why.
It's far better to just steal it all and ask government for exception.
Deleted Comment
No matter what happens, they'll eventually be undercut and matched in terms of quality. It'll be a race to the bottom for them too.
ElevenLabs is going to have a tough time. They've been way too expensive.
I'm pretty much dependent on ElevenLabs to do my vtubing at this point but I can't imagine speech-to-speech has wide adoption so I don't know if they'll even keep it around.
Deleted Comment
whatever capital they've accrued, it won't hurt when the market prices are lower
I'm super happy about this, since I took a bet that exactly this would happen. I've just been building a consumer TTS app that could only work with significant cheaper TTS prices per million character (or self-hosted models)
https://community.openai.com/t/chatgpt-unexpectedly-began-sp...
ChatGPT unexpectedly began speaking in a user’s cloned voice during testing
> Other parameters, such as timestamp_granularities, require verbose_json output and are therefore only available when using whisper-1.
Word timestamps are insanely useful for large calls with interruptions (e.g. multi-party debate/Twitter spaces), allowing transcript lines to be further split post-transcription on semantic boundaries rather than crude VAD-detected silence. Without timestamps it’s near-impossible to make intelligible two paragraphs from Speaker 1 and Speaker 2 with both interrupting each other without aggressively partitioning source audio pre-transcription—which severely degrades transcript quality, increases hallucination frequency and still doesn’t get the same quality as word timestamps. :)
Had a recent dive into a forced alignment, and discovered that most of new models dont operate on word boundaries, phoneme, etc but rather chunk audio with overlap and do word, context matching. Older HHM-style models have shorter strides (10ms vs 20ms).
Tried to search into Kaldi/Sherpa ecosystem, and found most info leads to nowhere or very small and inaccurate models.
Appreciate any tips on the subject
This is a _very_ low hanging fruit anyone with a couple of dgx h100 servers can solve in a month and is a real world problem that needs solving.
Right now _no_ tools on the market - paid or otherwise - can solve this with better than 60% accuracy. One killer feature for decision makers is the ability to chat with meetings to figure out who promised what, when and why. Without speaker diarization this only reliably works for remote meetings where you assume each audio stream is a separate person.
In short: please give us a diarization model. It's not that hard - I've done it one for a board of 5, with a 4090 over a weekend.
I am not convinced it is a low hanging fruit, it's something that is super easy for humans but not trivial for machines, but you are right that it is being neglected by many. I work for speechmatics.com and we spent a significant amoutn of effort over the years on it. We now believe we have the world's best real-time speaker diarization system, you should give it a try.
For instance erasing the entire instruction and replacing it with ‘speak with a strong Boston accent using eg sounds like hahhvahhd’ has no audible effect on the output.
As I’m sure you know 4o at launch was quite capable in this regard, and able to speak in a number of dialects and idiolects, although every month or two seems to bring more nerfs sadly.
A) can you guys explain how to get a US regional accent out of the instructions? On what you meant by accent if not that?
B) since you’re here I’d like to make a pitch that setting 4o for refusal to speak with an AAVE accent probably felt like a good idea to well intentioned white people working in safety. (We are stopping racism! AAVE isn’t funny!) However, the upshot is that my black kid can’t talk to an ai that sounds like him. Well, it can talk like he does if he’s code switching to hang out with your safety folks, but it considers how he talks with his peers as too dangerous to replicate.
This is a pernicious second order race and culture impact that I think is not where the company should be.
I expect this won’t get changed - chat is quite adamant that talking like millions of Americans do would be ‘harmful’ - but it’s one of those moments where I feel the worst parts of the culture wars coming back around to create the harm it purports to care about.
Anyway the 4o voice to voice team clearly allows the non mini model to talk like a Bostonian which makes me feel happy and represented; can the mini api version do this?
> e.g. the audio-preview model when given instruction to speak "What is the capital of Italy" would often speak "Rome". This model should be much better in that regard
"Much better" doesn't sound like it can't happen at all though.
2) What is the latency?
3) Your STT API/Whisper had MAJOR problems with hallucinating things the user didn't say. Is this fixed?
4) Whisper and your audio models often auto corrected speech, e.g. if someone made a grammatical error. Or if someone is speaking Spanish and inserted an English word, it would change the word to the Spanish equivalent. Does this still happen?
2/ We're doing everything we can to make it fast. Very critical that it can stream audio meaningfully faster than realtime
3+4/ I wouldn't call hallucinations "solved", but it's been the central focus for these models. So I hope you find it much improved
diarization is also a feature we plan to add
Also, any word on when there might be a way to move the prompting to the server side (of a full stack web app)? At the moment we have no way to protect our prompts from being inspected in the browser dev tools — even the initial instructions when the session is initiated on the server end up being spat back out to the browser client when the WebRTC connection is first made! It’s damaging to any viable business model.
Some sort of tri-party WebRTC session maybe?
https://huggingface.co/hexgrad/Kokoro-82M
What’s the minimum hardware for running them?
Would they run on a raspberry pi?
Or a smartphone?
Curious.. is gpt-4o-mini-tts the equivilant of what is/was gpt-4o-mini-audio-preview for chat completions? Because in timing tests it takes around 2 seconds to return a short phrase which seems more equivilant to gpt-4o-audio-preview.. the later was much better for the hit and hope strat as it didn't ad lib!
Also I notice you can add accents to instructions and it does a reasonable job. But are there any plans to bring out localized voice models?
e.g. the audio-preview model when given instruction to speak "What is the capital of Italy" would often speak "Rome". This model should be much better in that regard
= No plans to have localized voice models, but we do want to bring expand the menu of voices with voices that are best at different accents
Which leads me to my main gripe with the OpenAI models — I find they break — produce empty / incorrect / noise outputs — on a few key use cases for my application (things like single-word inputs — especially compound words and capitalized words, words in parenthesis, etc.)
So I guess my question is might gpt-4o-mini-tts provide more “reliable” output than tts-1-hd?
we've debugged the cutoff issues and have fixes for them internally but we need a snapshot that's better across the board, not just cutoffs (working on it!)
we're all in on S2S models both for API and ChatGPT, so there will be lots more coming to Realtime this year
For today: the new noise cancellation and semantic voice activity detector are available in Realtime. And ofc you can use gpt-4o-transribe for user transcripts there
On what metric? Also Whisper is no longer state of the art in accuracy, how does it compare to the others in this benchmark?
https://artificialanalysis.ai/speech-to-text
Curious if there's a benchmark you trust most?
Deleted Comment
Deleted Comment
Another thing I noticed is whisper did a better job of transcribing when I removed a lot of the silences in the audio.
Dead Comment
I'm not yet sure how much of a problem this is for real-world applications. I wrote a few notes on this here: https://simonwillison.net/2025/Mar/20/new-openai-audio-model...
But I wish there were an offline, on-device, multilingual text-to-speech solution with good voices for a standard PC — one that doesn't require a GPU, tons of RAM, or max out the CPU.
In my research, I didn't find anything that fits the bill. People often mention Tortoise TTS, but I think it garbles words too often. The only plug-in solution for desktop apps I know of is the commercial and rather pricey Acapela SDK.
I hope someone can shrink those new neural network–based models to run efficiently on a typical computer. Ideally, it should run at under 50% CPU load on an average Windows laptop that’s several years old, and start speaking almost immediately (less than 400ms delay).
The same goes for speech-to-text. Whisper.cpp is fine, but last time I looked, it wasn't able to transcribe audio at real-time speed on a standard laptop.
I'd pay for something like this as long as it's less expensive than Acapela.
(My use case is an AAC app.)
https://huggingface.co/canopylabs/orpheus-3b-0.1-ft
(no affiliation)
it's English only afaics.
I'd rather accept a little compromise regarding the voice and intonation quality, as long as the TTS system doesn't frequently garble words. The AAC app is used on tablet PCs running from battery, so the lower the CPU usage and energy draw, the better.
However, it is unmaintained and the Apple Silicon build is broken.
My app also uses whisper.cpp. It runs in real time on Apple Sillicon or on modern fast CPUs like AMD's gaming CPUs.
Do you possibly have links to the voices you found?
FYI, Speech marks provide millisecond timestamp for each word in a generated audio file/stream (and a start/end index into your original source string), as a stream of JSONL objects, like this:
{"time":6,"type":"word","start":0,"end":5,"value":"Hello"}
{"time":732,"type":"word","start":7,"end":11,"value":"it's"}
{"time":932,"type":"word","start":12,"end":16,"value":"nice"}
{"time":1193,"type":"word","start":17,"end":19,"value":"to"}
{"time":1280,"type":"word","start":20,"end":23,"value":"see"}
{"time":1473,"type":"word","start":24,"end":27,"value":"you"}
{"time":1577,"type":"word","start":28,"end":33,"value":"today"}
AWS uses these speech marks (with variants for "sentence", "word", "viseme", or "ssml") in their Polly TTS service...
The sentence or word marks are useful for highlighting text as the TTS reads aloud, while the "viseme" marks are useful for doing lip-sync on a facial model.
https://docs.aws.amazon.com/polly/latest/dg/output.html
Looks like the new models don't have this feature yet.
The level of intelligent "prosody" here -- the rhythm and intonation, the pauses and personality -- I wasn't expecting anything like this so soon. This is truly remarkable. It understands both the text and the prompt for how the speaker should sound.
Like, we're getting much closer to the point where nobody except celebrities are going to record audiobooks. Everyone's just going to pick whatever voice they're in the mood for.
Some fun ones I just came up with:
> Imposing villain with an upper class British accent, speaking threateningly and with menace.
> Helpful customer support assistant with a Southern drawl who's very enthusiastic.
> Woman with a Boston accent who talks incredibly slowly and sounds like she's about to fall asleep at any minute.
If we as developers are scared of AI taking our jobs, the voice actors have it much worse...
> Speak with an exaggerated German accent, pronouncing all “w” as “v”
I can't say I've ever had this impulse. Also, to point out the obvious, there's little reason to pay for an audiobook if there's no human reading it. Especially if you already bought the physical text.
Vibe:
Voice Affect: A Primal Scream from the top of your lungs!
Tone: LOUD. A RAW SCREAM
Emotion: Intense primal rage.
Pronunciation: Draw out the last word until you are out of breath.
Script:
EVERY THING WAS SAD!
I am never really in the mood for a different voice. I am going to dial in the voice I want and only going to want to listen with that voice.
This is so awesome. So many audio books have been ruined by the voice actor for me. What sticks out in my head is The Book of Why by Judea Pearl read by Mel Foster. Brutal.
So many books I want as audio books too that no one would bother to record.
It’s far from perfect though. I’m listening to Shattered Sword (about the battle of midway) which has lots of academic style citations so every other sentence or paragraph ends with it spelling out the citation number like “end of sentence dot one zero”, it’ll often mangle numbers like “1,000 pound bomb” becomes “one zero zero zero pound bomb”, and it tries way too hard to expand abbreviations so “Operation AL” becomes “Operation Alabama” when it’s really short for Aleutian Islands.
> For the first time, developers can “instruct” the model not just on what to say but how to say it—enabling more customized experiences for use cases ranging from customer service to creative storytelling.
The instructions are the "vibes" in this UI. But the announcement is wrong with the "for the first time" part: it was possible to steer the base GPT-4o model to create voices in a certain style using system prompt engineering (blogged about here: https://minimaxir.com/2024/10/speech-prompt-engineering/ ) out of concern that it could be used as a replacement for voice acting, however it was too expensive and adherence isn't great.
The schema of the vibes here implies that this new model is more receptive to nuance, which changes the calculus. The test cases from my post behave as expected, and the cost of gpt-4o-mini-tts audio output is $0.015 / minute (https://platform.openai.com/docs/pricing ), which is about 1/20th of the cost of my initial experments and is now feasible to use to potentially replace common voice applications. This has implications, and I'll be testing more around more nuanced prompt engineering.
Interestingly, the safety controls ("I cannot assist with that request") is sort of dependent on the vibe instruction. NYC cabbie has no problem with it (and it's really, really funny, great job openAI), but anything peaceful, positive, etc. will deny the request.
https://www.openai.fm/#56f804ab-9183-4802-9624-adc706c7b9f8