Access powerful AI models to transcribe and understand speech via a simple API. Try our no-code playground for free 👉 https://xitter.cf/t.co/YPCK9mqDG6
Joined October 2017
- Tweets2.8K
- Following415
- Followers45.9K
- Likes4.1K
Our Voice AI is moving into a haunted house.🎃👻
Join us for a haunted house happy hour and wander among life-size animatronics wired-up with AssemblyAI-powered voice agents.
Chat with them, collect candy along the way, and enjoy drinks, bites, and great company with fellow Voice AI builders.
Costumes are encouraged. Screams are optional.
Oct 29, 5–8 PM PT
luma.com/assemb-42oc
Thursday, October 22nd at 11am PT: a live first listen of our upcoming Text-to-Speech model with Tom Shapland, Head of New Products.
AssemblyAI-TTS-1 is our new Text-to-Speech model for voice agents. It’s built for conversation, with the natural pacing and self-corrections real callers expect.
Register: assemblyai.zoom.us/webinar/r…
City councils decide on real estate development and rezoning in meetings almost nobody watches.
Today @myhamletinc & @subes01 launched 10,000+ profiles, available for residents and journalists, powered by AssemblyAI Universal-3.5 Pro and speaker diarization.
Read the full story: assemblyai.com/customers/ham…
What's inside:
- One WebSocket for Speech-to-Text, LLM and voice
- Short-lived tokens, so your API key stays out of the browser
- Far-field Voice Focus for laptop and room mics
- Tool calls, up to leaving the meeting
Code: github.com/recallai/ai-voice…
Docs: assemblyai.com/docs/voice-ag…
Want an AI voice agent that joins your Zoom, Meet or Teams call and answers questions out loud?
@recallai just published the full build. AssemblyAI's Voice Agent API listens, thinks and speaks. Recall.ai's Output Media puts it in the meeting.
recall.ai/blog/how-to-build-…
@covaldev's latest streaming Speech-to-Text benchmark:
Universal-3.6 Pro has the lowest WER of any model tested.
Median time to final transcript dropped about a third this week:
3.6 Pro: 153 → 106 ms
3.5 Pro: 156 → 108 ms
Thanks to the Coval team for running these benchmarks in public every week.
Gemini 4 Argon is live on LLM Gateway.
On Voice Standard Bench it posted the top dictation score of 50+ models (99.5) and returned faster than Gemini 3.8 Flash at every transcript length.
Model string: gemini-4-argon
Full results: assemblyai.com/blog/gemini-4…
We are open sourcing Voice Standard Bench: 50+ LLMs evaluated over voice transcripts that matter!
In our Results GPT-6 Luna became go to for the value amd cost, while Qwen had the lowest latency
Try yourself on LLM Gateway the best for your use case!
AssemblyAI retweeted
message for those who think latency matters more than voice models accuracy!
"The difference between the almost right word and the right word is really a large matter—'tis the difference between the lightning bug and the lightning.
~Mark Twain
Universal-3.6 Pro is now on @OpenRouter, served through our Sync API.
One request in, one finished transcript back. No job to poll, no webhook, no WebSocket.
Already running your LLM calls through OpenRouter? Your Speech-to-Text can go on the same key.
Universal-3.6 Pro from @AssemblyAI is now live on OpenRouter
Speech-to-text through AssemblyAI's Sync API. Compared with Universal-3.5 Pro:
- 30% fewer errors on short conversational utterances
- 12% fewer errors on average across supported languages
- Better recognition and formatting of phone numbers, addresses, emails, and codes
- 14 new languages: Korean, Catalan, Galician, Russian, Romanian, Estonian, Persian, Afrikaans, Marathi, Cantonese, Zulu, Urdu, Xhosa, and Norwegian Nynorsk
Try it now at only $0.30/hr
openrouter.ai/assemblyai/uni…
AssemblyAI retweeted
models great at coding aren't always great at voice.
we are open sourcing Voice Standard Bench: 50+ LLMs evaluated over voice transcripts that matter!
what model you would use?
you can compare and try it yourself with LLMGateway on AssemblyAI
LLM Gateway now has 50+ models. Voice Standard Bench ranks every one on real Speech-to-Text output, with quality, speed and cost side by side.
Up to 15 minutes of audio, a model within two quality points of the leader costs 50 to 85× less.
assemblyai.com/voice-standar…
All 50+ models in the run are available through LLM Gateway with your existing AssemblyAI API key, and the benchmark is open source, so you can run it on your own transcripts. How we test, and where to start:
assemblyai.com/blog/voice-st…
Speed matters for voice agents, and open models are the fastest in the run on short transcripts: on 30-second and 2-minute inputs the fastest answer in under a second (median). Every failure counts against the quality score too: wrong format, refusal, timeout, error.
The main finding: for most voice tasks you don't need to pay frontier-model prices. On transcripts up to 15 minutes, a model within two quality points of the top scorer costs 50 to 85× less per hour of audio. On hour-long transcripts the gap closes to about 3×.
LLM Gateway now has 50+ models. Today we're releasing Voice Standard Bench, which scores all of them on real Speech-to-Text output across four voice tasks, on transcripts from 30 seconds to an hour. Quality, speed and cost are reported side by side.
assemblyai.com/voice-standar…
Universal-3.6 Pro is now live on Bolna (YC F25).
Entity-aware endpointing so callers finish their account number before the agent jumps in. 32 languages, including Hindi, Urdu and Marathi.
Free for all Bolna users 9 to 16 October.
AssemblyAI retweeted
@AssemblyAI Universal-3.6 Pro is now live on @bolna_dev!
It is fine-tuned on human agent conversations, and that shows in the details your agents act on. Entity-aware endpointing helps the agent avoid cutting off a caller halfway through an account number.
Accuracy has improved, and it now supports 32 languages, including Hindi, Urdu, Marathi and more. Phone numbers, addresses, emails and codes are recognised and formatted according to how your agents need them.
PS: It’s free for all Bolna users from 9th to 16th of October. We’re excited to have you try it, so let us know how your agents perform.
Who knew a nostalgic childhood toy could become a voice agent? 🧸🎤
For SF Tech Week, we brought the voice Al community together to give a Furby a voice upgrade using AssemblyAl's Voice Agent API.
From swear jar Furbies to karaoke sing-along Furbies, we loved seeing all the creative (and hilarious) ideas come to life!
Huge thanks to everyone who joined us, built something fun, and took home their very own Furby pal
