Deepgram · 索引內容

ChangelogSubscribe via RSSSelect filtersSep 18, 2026Correction: Browser Agent SDK has no client-side VAD The Browser Agent SDK never shipped client-side Silero voice activity detection. Documentation published with the SDK described it as an available feature, and that documentation was wrong. The following did not exist in any released package: The vad option on MicrophoneOptions in @deepgram/agents. The speechThreshold and silenceThreshold VAD options. The speech-start and speech-end events on AgentMicrophone. The vad field in WidgetConfig and the widget’s VAD configuration section. The @ricky0123/vad-web and onnxruntime-web peer dependencies. Uninstall them if you added them for VAD. Voice AgentCorrection: token minting uses ttl_seconds The server-side token-minting example on the Browser Agent SDK overview posted {"ttl": 30} to POST /v1/auth/grant. The field is named ttl_seconds: body: JSON.stringify({ ttl_seconds: 30 }), /v1/auth/grant accepts unrecognized fields, returns 200, and issues a token with the 30-second default, so a request built from the old example raised no error while ignoring the requested lifetime. If you copied that example and increased the number, your tok…

Built for the language of pharmaPower accurate, reliable, and secure pharmaceutical voice applications with speech-to-text optimized to transcribe drug names, pharmaceutical terminology, and medical vocabulary.Try in API PlaygroundSign up for freeLoading video...The leader in pharmaceutical speech recognitionGeneral-purpose clinical STT models routinely mistranscribe drug names and pharmaceutical terminology, introducing errors that can affect patient care and critical pharmaceutical workflows. Nova-3 Pharma is specifically designed to accurately transcribe the terminology those workflows depend on.#1 in pharmaceutical terminology accuracyAccurately transcribes drug names, pharmaceutical terminology, and medical vocabulary while maintaining the same high-quality transcription performance of Nova-3. It ranks #1 in drug name recognition across both batch and streaming among the models evaluated.Designed specifically for pharmaceutical workflowsBuild pharmacy IVRs, prescription refill automation, member services and contact center voice agents, and clinical documentation on a single speech-to-text platform.Seamless deploymentDeploy using model=nova-3-pharma with no infrastructure chan…

Listen to article03:42Table of ContentsIn pharmaceutical workflows, getting a medication name wrong isn’t just another transcription error. It can introduce incorrect information into critical workflows, from prescription refill automation and pharmacy IVRs to clinical documentation. A transcript can score highly on overall accuracy, but an error on a medication name can carry far greater consequences than errors elsewhere in the transcript.That’s why we built Nova-3 Pharma, the first speech-to-text model purpose-built for the pharmaceutical industry. Nova-3 Pharma is specifically designed to accurately transcribe drug names, pharmaceutical terminology, and medical vocabulary while maintaining the same high-quality transcription performance Deepgram customers expect from Nova-3. In our latest competitive benchmarks, Nova-3 Pharma ranks #1 in drug name recognition and #1 in overall transcription accuracy across both batch and streaming.With Nova-3 Pharma, developers can build pharmacy IVRs, prescription pickup and refill automation, member services and contact center voice agents, and clinical documentation applications where drug name accuracy is critical.It’s available now for bot…

{"@context":"https://schema.org","@type":"Product","name":"Deepgram Voice AI Platform Pricing","description":"Usage-based pricing for Deepgram speech-to-text, text-to-speech, voice agent, and audio intelligence products.","brand":{"@type":"Brand","name":"Deepgram"},"url":"https://deepgram.com/pricing","category":"Voice AI platform","offers":[{"@type":"Offer","name":"Deepgram Voice AI Platform Pricing - Streaming - Flux English - Pay As You Go","price":"0.0065","priceCurrency":"USD","availability":"https://schema.org/InStock","url":"https://deepgram.com/pricing"},{"@type":"Offer","name":"Deepgram Voice AI Platform Pricing - Streaming - Flux English - Growth","price":"0.0057","priceCurrency":"USD","availability":"https://schema.org/InStock","url":"https://deepgram.com/pricing"},{"@type":"Offer","name":"Deepgram Voice AI Platform Pricing - Streaming - Flux Multilingual - Pay As You Go","price":"0.0078","priceCurrency":"USD","availability":"https://schema.org/InStock","url":"https://deepgram.com/pricing"},{"@type":"Offer","name":"Deepgram Voice AI Platform Pricing - Streaming - Flux Multilingual - Growth","price":"0.0068","priceCurrency":"USD","availability":"https://schema.org/InStock…

Deepgram Speak '26San Francisco | October 29Reserve your spot49d : 01h : 02m : 35sThe future is talking. Turn it up.Deepgram Speak brings the people building voice AI together for one day, in one room. Come curious. Come ready to compare notes. There's a lot to talk about.What you can expect:Live product moments and hands-on demos.Keynotes and fireside chats on where voice AI is headed.Real conversation with the people building and shipping voice AI, not sales pitches.Reserve your spotOur event sponsorsBecome a sponsorSave the dateWhenThurs, October 298:00 AM – 6:00 PMWhereThe AviaryThe Aviary (opens in a new tab)135 Fourth St, Ste. 4000 San Francisco, CAThe future is talking. Turn it upReserve your spot

Listen to article11:24Table of ContentsSpeaker diarization is the process of automatically identifying and separating individual speakers within an audio stream. When applied to automatic speech recognition (ASR) transcripts, speaker diarization labels each utterance with its corresponding speaker, transforming a continuous block of text into a structured, readable conversation format.Each speaker is identified by their unique audio characteristics—including vocal pitch, cadence, and acoustic patterns—and their utterances are grouped together under consistent labels. This capability is also referred to as speaker labels, speaker segmentation, or speaker change detection.Speaker diarization answers the fundamental question: "Who spoke when?" Understanding what speaker diarization is and how it works is essential for building voice-enabled applications that require speaker attribution.Example with speaker diarization:Without speaker diarization:My name is Beth, and I will be assisting you today. How are you doing? Not too bad. How are you today? I'm doing well. Thank you. May I please have your name? My name is Blake.With speaker diarization:[Speaker:0] My name is Beth, and I will be…

Listen to article11:20Table of ContentsMeta Title: Voice AI latency: why pauses read as hesitationMeta-description: Learn how voice AI latency affects user perception, why silence reads as incompetence, and how to set latency budgets by deployment context in 2026.URL Slug: voice-ai-latency-why-pauses-read-as-hesitationVoice AI latency: why pauses read as hesitationIn a Deepgram survey of 40 developers building voice agents, 61% reported noticeable to major flow issues. Flow breaks when a reply arrives nearly a second after the caller stops talking, which reads as hesitation even when every word is right.The person on the call draws their own conclusion from your delay. Voice AI latency becomes a perception problem before it becomes a metrics problem. And the meaning gets worse as the silence grows.End-of-turn detection, the language model, and speech synthesis each consume part of the delay you can afford, so no single component owns the total. Latency budgets differ by deployment context: phone IVR, in-app assistant, drive-through ordering, and live captioning each need a separate target.In this guide, you'll see what a caller reads into a pause, and when your agent should start t…

Deepgram vs. AssemblyAISee why 200,000+ developers prefer Deepgram for streaming and batch STT.Deepgram offers the most accurate streaming STT APIDeepgram far outperforms Assembly AI in performance benchmarksDeepgram offers on-premises and cloud APIs Start for FreeBased on 250+ reviewsTrusted by industry leadersDeepgram beats AssemblyAIBuild with enterprise-grade speech recognition that's faster, more accurate, and affordable. No compromises.Flexible deploymentChoose cloud, on-premises, or private cloud to securely manage voice and transcription data with Kubernetes, Docker, and pre-built VM support for easy setup in any environment.Custom model trainingDeepgram offers tailored ASR models optimized with customer-specific data, ideal for industries with specialized jargon, accents, or unique speech patterns.Enterprise securityProtect customer data privacy and ensure regulatory compliance with HIPAA-compliant transcription.Innovation leader in Voice AIDeepgram's deep learning models are optimized for speech data and trained on diverse datasets, delivering industry-leading performance in pre-recorded and real-time transcription.Fast and accurate transcriptionDeepgram's speech-to-text …

Listen to article02:41Table of ContentsSo, you want to run the OpenAI Whisper tool on your machine? You can load it from the OpenAI Github repository to get up and going!SetupYou'll need python on your machine, at least version 3.7. Let's set up a virtual environment with venv (or conda or the like) if you want to isolate these experiments from other work.Bashmkdir whisper cd whisper python3 -m venv venv source venv/bin/activate # always a good idea to make sure pip is up-to-date pip3 install --upgrade pipNext, install a clone of the Whisper package and its dependencies (torch, numpy, transformers, tqdm, more-itertools, and ffmpeg-python) into your python environment.Bashpip3 install git+https://github.com/openai/whisper.gitEspecially if it's pulling torch for the first time, this may take a little while. The repository documentation advises that if you get errors building the wheel for tokenizers, you may also need to install rust. You'll also need ffmpeg - installation depends on your platform. Here are some examples:Bash# on Ubuntu or Debian sudo apt update && sudo apt install ffmpeg # on Arch Linux sudo pacman -S ffmpeg # on MacOS using Homebrew (https://brew.sh/) brew install …

Listen to article06:04Table of ContentsSo, you've probably heard about OpenAI's Whisper model; if not, it's an open-source automatic speech recognition (ASR) model – a fancy way of saying "speech-to-text" or just "speech recognition." What makes Whisper particularly interesting is that it works with multiple languages (at the time of writing, it supports 99 languages) and also supports translation into English. It also has a surprisingly low word error rate (WER) out-of-the-box.Whisper makes it pretty easy to invoke at the command line, as a CLI:Bash$ curl -sSfLO https://static.deepgram.com/example/tenant_of_wildfell_hall.mp3 $ whisper tenant_of_wildfell_hall.mp3 Detecting language using up to the first 30 seconds. Use `--language` to specify the language Detected language: English [00:00.000 --> 00:07.000] On entering the parlour, we found that Honoured Lady seated in her armchair at the fireside, [00:07.000 --> 00:27.000] working away at her knitting.And here's an example of its language detection at work:Bash$ curl -sSfLO https://static.deepgram.com/example/el_caso_leavenworth.mp3 $ whisper el_caso_leavenworth.mp3 Detecting language using up to the first 30 seconds. Use `--langu…

Listen to article03:14Table of ContentsNuance recently acquired Saykara, a mobile speech recognition technology provider to expand their medical transcription business. This acquisition is one of many major investments and acquisitions in the Natural Language Processing (NLP) and Customer Service space that we are monitoring at Deepgram. Due to the recent acquisition of Saykara, we thought it would be a good time to review Nuance speech recognition capabilities and why customers creating real-time experiences should consider alternatives. Enterprise audio is more nuanced than it may seem - pun intended!Leader of previous speech tech solutionsNuance is a great brand in the speech recognition business and has been around for over 30 years. They have gobbled up smaller speech recognition businesses, including Saykara just recently to expand their medical transcription business. If you need speech to text transcription especially in the medical setting, they will always be on the list to evaluate. To be honest, our most recent survey indicates good satisfaction with Nuance.Core architecture has remained unchangedNuance does a good job at speech to text transcription using their 1970's …

Listen to article01:56Table of ContentsDeepgram is excited to continue executing the vision of having everyone heard and understood through speech recognition. To that end, we are announcing five new languages and three new use case speech models. In addition, we have improved our Hindi and Spanish language models and our Conversational AI and Meetings models. These improvements mean better accuracy for your transcription applications. Through transfer learning backed by our End-to-End Deep Learning architecture and the process of creating conversational audio datasets for training, we have been able to achieve very high accuracy on our new speech models. Our new models are trained on real conversations, not Libraspeech type datasets of people reading audiobooks because no one talks like they are reading audiobooks. This allows our speech models to better transcribe jargon, accents, dialects, code-switching (switching from Hindi to English words), terminology (ROI, EBITDA), and alphanumerics. This means our models provide better real-world transcription accuracy than standard language or general one-size-fits-all speech models. Improved Production Language ModelsHindi (improved)Spa…

Listen to article03:24Table of ContentsBack in January, we supported Hack Cambridge - a 24-hour student hackathon. The team behind AutoBubble wanted to see if they could improve the display of captions for online videos. I sat down with Andy Zhou, Conall Moss, Dan Wendon-Blixrud, and Lochlann-B Baker to ask them about their project.The Project"There were a lot of challenges and prompts at Hack Cambridge, but the Deepgram challenge was both the most flexible and the coolest" explains Conall. "We knew we were going to use it but then had to think of an idea."Dan continues: "A lot of speaker communication comes through facial expressions, and while closed captions are super useful, they are generally in a fixed position. We wanted to build a project which allows for captioning while allowing the depth of expression."With that, AutoBubble was born. It is a Chrome extension that uses facial recognition and Deepgram's Speech Recognition API to place captions next to a speaker's face in a YouTube video.First-Time HackersThe team behind AutoBubble are all first-year Computer Science students at the University of Cambridge and, amazingly, were taking part in their very first hackathon. All …

Listen to article01:52Table of ContentsWe're pleased to announce that we've recently received recognition from two different venues. First, we're now number 1 on G2 in the Voice Recognition Software category, as well as being highlighted as a High Performer in G2's most recent quarterly report. Second, we won a Silver Stevie Award in the Sales or Customer Service Solutions Technology Partner of the Year category. One of the driving forces for both of these recognitions is our focus on customer service and success. If you've ever tried to use Big Tech's automatic speech recognition (ASR) solutions, you know that if you run into trouble or need help from an actual human, you're pretty much out of luck. You just have to hope that whatever your problem is clearly documented.Customer Success is at the Heart of DeepgramAt Deepgram, however, we aspire to go beyond these kinds of transactional business relationships with our customers and instead aim to build partnerships. We want to ensure that all of our customers are achieving their desired outcomes and seeing the value of their partnership with Deepgram. Our Customer Success team works hand-in-hand with each customer to have a clear vi…

Listen to article16:26Table of ContentsRecently I wrote about a project I did to help me learn Pinia, Vue 3's new official state management system. I built a basic to-do list app:It dawned on me that a fun way to jazz up this project would be to use Deepgram to make the app voice-powered so that a user can speak commands to add, delete, or check-off items on the list.I'm inspired by my colleague Bekah's series about updating portfolio projects. A voice-based to-do list app would be a lot more interesting than a regular to-do list app!Build a To-Do List App With Vue 3, Pinia, and Deepgram (SERIES)Build a To-do List App with Pinia and Vue 3Build a Voice Controlled To-Do List App with Deepgram and Vue 3The project I originally did can be found in this repo, and the accompanying blog post is here. Check it out to build the standard to-do list project with Vue 3 and Pinia.In this iteration of the project, I'll continue to use Pinia to manage global state, but I'll add Deepgram so I can use Deepgram's speech-to-text API to help me power the voice-control feature. If you want to build this voice-control feature along with me, I've created a starting branch here.Deepgram Live Streaming Log…

Enterprise Voice AI on AWS, integrated into AWS ConnectAccurate, real-time transcription for agent assist, CX workflows, and automation, all on AWS.Real-time transcription, automated workflows, whatever you're building, you can now do it with models that actually perform. And your voice agents will sound human, not robotic.Seamless integration with existing Connect and Lex workflows, no hacks, no heavy liftingUltra-low latency for natural, real-time conversationsState-of-the-art accuracy across noisy environments and diverse accentsScalability to support the largest enterprise deploymentsTrusted byTrusted by

Super-Charge Your CX With Accurate Speech RecognitionDeepgram is the most accurate, fastest, scalable, and affordable Automatic Speech Recognition (ASR) solution for enterprises and software companies.Talk To An ExpertWhat Is DeepgramLoading video...Trusted by the world’s top Enterprises and StartupsSpeech AI Designed with Your Use Case in MindUnleash the potential of Deepgram's cutting-edge speech recognition API across diverse applications. From enhancing accessibility in real-time transcriptions to powering voice assistants for effortless interactions, Deepgram delivers unparalleled performance for every use case.Contact CentersTranscribe customer interactions into actionable data and analyze customer sentiment, detect key topics to enable agents in real time.Conversational AIPower AI virtual assistants with robust understanding capabilities for more intuitive conversational AI experiences.Speech AnalyticsTranscribe conversational data for analysis to surface relevant insights, monitor regulatory compliance, QA, and more.Media TranscriptionCaption, summarize, and analyze podcasts and videos affordably and efficiently to streamline production workflows.E-learningEnhance online le…

Listen to article13:04Table of ContentsIf you’d rather watch than read, check out the video version of this blog here:Loading video...As they say on AI Twitter: “AI won’t replace you, but a person who knows how to use AI will.”It turns out, this phrase doesn’t just apply to writers, SEO managers, and lawyers. It applies to software engineers as well.That’s right. The very engineers who forged AI from the fiery depths of linear algebra are now (at least tacitly) required to use their creation… lest they fall behind in the technological innovation race.In other words, having an AI assistant help you code boosts your productivity immensely. Not only does the AI write quick, trivial functions for you, but such models can also help you debug and comment your code near-instantly.That being said, there’s one problem: Numerous AI coding assistants exist, and we don’t know which one is the best. It’d be rather difficult and time-consuming to test them all. There’s just too many. Not to mention, we need to take into consideration the cost of each one—both in terms of money and onboarding-time. After all, the learning curve for these AI tools should be reasonably shallow. The more intuitive a…

Listen to article08:45Table of ContentsWhen it comes to learning a new topic, everyone’s wired differently; some are perfectly content to squat in theory-land indefinitely, some enjoy grappling with theory for a few rounds before hobbling together a pragmatic application, and some prefer learning Bob Ross style—which is to say watching someone do the thing you’re interested in learning and following along.​For this last option, YouTube is a wonderful resource. Particularly for newcomers to the field, when a new machine learning framework arrives, it can be motivating to see someone else paint out for you what that framework can do before you slog through its documentation.​Released in October 2022 and rapidly evolving since, the LangChain framework is designed for hacking together applications that connect large language models (LLMs) with other tools at fairly high levels of abstraction, ideally giving rise to more complex functionality than an LLM can offer alone. Though they make it powerful, LangChain’s numerous components—schemas, models, prompts, indexes, memory, chains, and agents—can make LangChain a lot to wrap your head around.​Thankfully, via videos filled with examples,…

Listen to article08:21Table of ContentsJust about everyone’s heard—and likely felt—the sentiment, “This meeting could’ve been an email,” but in spite of that, in-person meetings and check-in conference calls are not going to stop being part of the business world any time soon. Fortunately, recent advancements in artificial intelligence technology are poised to revolutionize how we retain information shared in meetings, as well as how effectively we can use that information once all participants go their separate ways.A good AI transcription tool should not only copy exactly what’s said aloud, but also be able to identify distinct speakers, extract important action items and assignments, collate information for easy review and retrieval later, and increase accessibility for people with hearing loss or anyone who may have trouble devoting their full attention to any particular meeting. Here are some of the top AI meeting transcription tools that are available in 2023, and some details about Deepgram’s efforts to push the boundaries of AI technology further than before.Understanding AI Transcription ToolsOne of the biggest benefits of AI transcription is also one of the most obvious: …

Listen to article01:07Table of ContentsThis blog post should be a quick one. Long story short, I have a database in Fauna that contains loads upon loads of URLs. And for a reason I shall not disclose—though you are welcome to playfully speculate—I had to delete quite a few URLs from this database.Here’s the problem: Fauna’s documentation sucks has room for improvement.So allow me to spare you the half-hour it took me to figure out how to delete multiple entries from a database in Fauna. Really all you have to do isGrab your `secret`, the name of your table, and your regex and plop them into the script below.Profit.I highly recommend printing the results first before deleting anything just to be sure that `results` contains the entries you indeed wish to delete. But beyond that, there’s not much else to do. The script below honestly boils down to initializing fauna and running a for-loop that deletes the entries that match the regex.Javascriptconst faunadb = require('faunadb') const q = faunadb.query const regex = /YOUR REGEX HERE/ const table = “table name here” const runner = async () => { const client = new faunadb.Client({ secret: 'YOUR_SECRET_HERE', }) const results = await cli…

Listen to article12:45Table of ContentsAn energy-threshold detector, scored against 30 hours of hand-annotated human conversation, fired on 45% of the backchannels and non-speech noise it heard. In casual talk, where backchannels are densest, that rate climbed to 52%.None of those triggers was a bid to speak, and every one of them would stop a voice agent mid-sentence. Turn detection separates a listener's "keep going" from a caller taking the floor. An energy threshold can't make that distinction.Your transcript already carries most of it. In this guide you'll find which tokens to look for, how long to wait after one, and what to log when your policy guesses wrong.Key takeawaysFour things decide whether your agent holds its ground or stops talking:A backchannel is a continuer that tells the speaker to keep going rather than a signal that the listener wants the floor.Energy-based barge-in can't separate the two, because both arrive as speech energy while your agent is talking.The specific backchanneling tokens (uh-huh, mhmm) land in your transcript once filler words are on; handling them is a lookup.The negative forms (uh-uh, nuh-uh, mm-mm) are the exception. They're two-syllable c…

Deepgram for StartupsBuild your voice AI product on the platform startups trust.Get the accuracy, speed, and reliability your product demands on a single voice AI platform. Apply to Deepgram for Startups and get up to $100,000 in credits over 12 months.Applicants must be:An AI builder or early-stage startupIn production or looking to launch in the next 6 monthsCommitted to actively building and engaging with our communityApply NowThose selected get exclusive access to:API CreditsUp to $100k Deepgram credits to build with Deepgram's powerful STT, TTS, Voice Agent capabilities. Credits must be used within 12 months.Community & NetworkingAccess invite-only events, engineering support groups, and the Deepgram startup community.99.5%Enterprise AccuracyIndustry-leading speech recognition across accents and domains.<200msUltra-Low LatencyReal-time streaming for live conversations and voice agents.EnterpriseSecuritySOC 2, PCI, GDPR, and HIPAA compliant.50+Global ScaleLanguages supported with consistent quality worldwide.Apply today!Create a Deepgram account to apply. Have your Project ID handy. Sign in / Create account — opens in a new tab, your progress stays here.How it works1. ApplyTell…

Deepgram x EWOR Startup ProgramDeepgram has partnered with EWOR to provide up to $100,000 in credits to power your voice AI product for 12 months.Apply NowDo more with voiceDeepgram is a comprehensive AI transcription foundation plus the understanding features you need to make your data readable and actionable by humans...or machines.Access to AI researchers and engineerswho will work hand-in-hand with you to transform your vision into reality.Up to $100k Deepgram creditsUse these credits to access all our powerful STT, TTS, Voice Agent, and Audio Intelligence APIs.Join a thriving voice AI startup communityGet invited to exclusive startup events, learn from industry experts, and connect with other builders.Topic summarizationAccurately identify, extract, and summarize conversational audio in your product.Join the programRequirements to join:Created a free Deepgram accountYou're a talented AI builder or an early-stage startup (<$10m raised)Committed to actively building and engaging with our communityIn production or looking to launch in the next 6 monthsApply nowDeepgram’s transcription is super fast and accurate. Their managed whisper API has also been huge accelerator for how qui…

Deepgram x AWS Startup ProgramDeepgram has partnered with AWS to provide up to $100,000 in credits to power your voice AI product for 12 months.Apply NowDo more with voiceDeepgram is a comprehensive AI transcription foundation plus the understanding features you need to make your data readable and actionable by humans…or machines.Access to AI researchers and engineerswho will work hand-in-hand with you to transform your vision into reality.Up to $100k Deepgram creditsUse these credits to access to all our powerful STT, TTS, Voice Agent, and Audio Intelligence APIs.Join a thriving voice AI startup communityGet invited to exclusive startup events, learn from industry experts, and connect with other builders.Topic summarizationAccurately identify, extract, and summarize conversational audio in your product.Join the programRequirements to join:Created a free Deepgram accountYou're a talented AI builder or an early-stage startup (<$10m raised)Committed to actively building and engaging with our communityIn production or looking to launch in the next 6 monthsApply nowDeepgram’s transcription is super fast and accurate. Their managed whisper API has also been huge accelerator for how quic…