Sold by: AssemblyAI
Deployed on AWS
AssemblyAI is the Voice AI infrastructure for developers building products that transcribe, understand, and act on speech. One API gives you state-of-the-art speech models for pre-recorded and real-time audio, serverless inference that scales from zero with no GPUs to manage, and built-in speech understanding across 99+ languages: speaker labels, language detection and code-switching, PII redaction, and LLM analysis. Voice agents, contact centers, AI notetakers, and healthcare scribes run on AssemblyAI, which processes more than 120 million conversations a week on AWS.
Start with $50 in usage credits during a 90-day free trial, then pay as you go with no commitment. Usage is billed through your AWS account and can count toward eligible AWS spend commitments. Volume pricing is available through private offers.
4.3
Overview
AssemblyAI is the Voice AI infrastructure layer for developers. We build the speech models, run the inference, and handle the work between raw audio and a usable result, so your team ships voice features instead of maintaining a speech stack.
WHY TEAMS BUILD ON ASSEMBLYAI
- Models built for real audio. Our speech models are trained on real conversations, accents, and noisy phone lines, not clean benchmark audio. They are most accurate where voice products break: names, account numbers, emails, addresses, and medical terms. We publish our benchmarks, including where we lose.
- Inference built for these models. Serverless and vertically integrated, with no GPUs to manage, no cold starts, and instant scale from zero. You pay only for the audio you process.
- Speech understanding built into the API. Language detection, code-switching, speaker diarization, contextual prompting, PII redaction, and LLM analysis run in the same call, so there is no glue code to write and maintain.
WHAT YOU CAN BUILD
- Voice agents that catch the caller's details and know when the caller is done talking
- AI notetakers and meeting intelligence with accurate speaker labels
- Contact center analytics, agent assist, and QA at the scale of every call
- Clinical documentation and AI scribes with Medical Mode
- Dictation and voice input that returns send-ready text
- Captioning, media search, and searchable audio archives
ONE API FOR THE WHOLE VOICE STACK
Transcribe
- Pre-recorded speech-to-text: our most accurate models for recorded audio and video, with 99+ languages, word-level timestamps, speaker labels, keyterms, and prompting
- Real-time speech-to-text: low-latency streaming with built-in turn detection, mid-sentence code-switching across 32 languages, and a model tuned for voice agents. First-party plugins for Pipecat and LiveKit
- Sync speech-to-text: a finished transcript from a single HTTP call for short audio, with no WebSocket to manage
Understand
- Speech Understanding: speaker identification by name, entity detection, sentiment, topic detection, translation, and custom formatting, priced per hour and stacked on the transcript
- Guardrails: PII redaction for text and audio, content moderation, and profanity filtering
- LLM Gateway: run frontier and open LLMs on your transcripts through one OpenAI-compatible API at provider rates, with automatic fallbacks
Interact
- Dictation API: one call returns cleaned-up, send-ready text alongside the verbatim transcript
- Text-to-Speech: AssemblyAI-TTS-1, trained on real conversational speech so voice agents sound natural and get names and numbers right
BUILT ON AWS
Our cloud API runs on AWS, with US and EU data residency endpoints. Buying through AWS Marketplace consolidates billing on your AWS invoice and can count toward eligible AWS spend commitments. Teams that need audio to stay inside their own AWS account can self-host our streaming, sync, and speaker diarization models in their own VPC, as containers on Amazon ECS or on Amazon SageMaker. Customers commonly pair AssemblyAI with Amazon Bedrock, Amazon Connect, Amazon S3, and AWS Lambda.
SECURITY AND COMPLIANCE
- SOC 2 Type 2, ISO 27001:2022, and PCI DSS v4.0
- HIPAA Business Associate Agreement, signable without a sales call
- GDPR, with US and EU data residency
- TLS 1.2+ in transit and AES encryption at rest
- Configurable retention down to one hour, deletion on demand, and opt-out from model training
TRUSTED AT SCALE
More than 3,000 companies build on AssemblyAI, including Zoom, Granola, Fireflies, Dovetail, CallRail, Aircall, and ClickUp. Siro cut customer complaints and support tickets by 90% after moving transcription to AssemblyAI, and Dovetail improved word error rate by 36%.
PRICING
Usage-based, billed per hour of audio, with no upfront commitment. Speech Understanding and Guardrails features are priced per hour and add to the base transcription rate. LLM Gateway is billed per token. New accounts get $50 in usage credits during a 90-day free trial. For committed-use discounts, request a private offer.
Highlights
- Accurate on the words that matter: names, numbers, emails, and medical terms. Our models lead our published voice-agent benchmark on entity accuracy, and prompting, key terms, and Medical Mode tune them to your domain.
- Infrastructure, not a wrapper: serverless inference built for our own models, so you scale from zero to millions of hours with no GPUs to manage, no cold starts, and unlimited concurrency on pre-recorded audio.
- Enterprise-ready on AWS: SOC 2 Type 2, ISO 27001, PCI DSS v4.0, a self-serve HIPAA BAA, US and EU data residency, and retention you control. Buy through Marketplace and apply spend to your AWS commitment.
Details
Sold by
Categories
Delivery method
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Trust center
Access real-time vendor security and compliance information through their Trust Center powered by Drata or Vanta. Review certifications and security standards before purchase.
Buyer guide
Gain valuable insights from real users who purchased this product, powered by PeerSpot.
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on actual usage, with charges varying according to how much you consume. Subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost/unit |
|---|---|---|
Universal-2 | Fast, intelligent async transcription with exceptional accuracy and unlimited concurrency | $0.15 |
SLAM-1 (deprecated) | Highest accuracy transcription powered by LLM intelligence | $0.27 |
Universal Streaming | Fast, accurate real-time transcription. Built-in turn detection and unlimited concurrency | $0.15 |
Keyterms Prompting (Universal Streaming) | Improve recognition accuracy for specific words and phrases | $0.04 |
Speaker Identification | Identify speakers by their actual names and roles | $0.02 |
Translation | Automatically convert your transcribed audio content from one language to another | $0.06 |
Custom Formatting | Ensure consistency through automatic, standardized formatting | $0.03 |
Entity Detection | Identify entities like person and company names, email addresses, dates, and locations | $0.08 |
Sentiment Analysis | Detect the sentiment of each sentence of speech spoken in your audio files | $0.02 |
Auto Chapters | Automatically generate a summary over time for audio and video files | $0.08 |
Dimensions summary
You pay per use, billed hourly for audio or session time. Core transcription models form the base: async (Universal-3.5 Pro, Universal-2, SLAM-1), streaming (Universal Streaming, Universal-3.5 Pro Realtime), plus Sync, Dictation, and Voice Agent APIs. Most other dimensions are add-ons that stack on a base model rate, including Speaker Diarization, Keyterms Prompting, Prompting, Medical Mode, and Voice Focus. Speech Understanding options (Translation, Sentiment Analysis, Summarization, Topic Detection, and more) and Guardrails (PII Redaction, Content Moderation, Profanity Filtering) also bill separately. LLM Gateway bills per token, varying by the model you choose.
Top-of-mind questions for buyers
Streaming bills on session duration, meaning the time the connection stays open, not the audio actually sent. Idle time between calls still counts. Close connections right when a call ends to avoid charges for open, unused sessions.
Add-ons stack additively on the base model rate per hour. For example, Universal-3.5 Pro plus Medical Mode plus standard Speaker Diarization adds each add-on rate to the base rate. The base model usually drives most of the cost, with add-ons increasing it.
Multichannel audio bills per channel. A one-hour stereo file with two channels counts as two hours of transcription. Each channel transcribes independently. Confirm channel count upfront and downmix if you do not need separate per-channel transcripts.
www.assemblyai.com
Helpful?
Vendor refund policy
All fees are non-refundable and non-cancellable except as required by law.
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Software as a Service (SaaS)
SaaS delivers cloud-based software applications directly to customers over the internet. You can access these applications through a subscription model. You will pay recurring monthly usage fees through your AWS bill, while AWS handles deployment and infrastructure management, ensuring scalability, reliability, and seamless integration with other AWS services.
Resources
Vendor resources
Support
Vendor support
24/7 support via chat at assemblyai.com and email at support@assemblyai.com .
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Enterprise-grade AI voice agents for contact centers, with real-time translation, voice identity transformation, and compliance-ready observability.
Smallest Atoms is a platform for deploying hyper-realistic AI agents that can talk to anyone on voice or text, in any language and any voice. Atoms act like AI employees that handle end-to-end workflows across support, sales, and operations, using your business context, tools, and data. They can automate conversations such as answering queries, collecting information, completing transactions, and triggering downstream actions, while continuously learning from mistakes. Atoms integrate into your stack via SDKs and APIs, making it easy for teams to embed real-time, natural conversations into web, mobile, and backend systems.
Omnichannel conversational AI agents that resolve customer enquiries across voice and chat, understanding every caller regardless of accent, dialect, or language. Built on Raven, PolyAI's proprietary customer-service LLM, and deployed by enterprises handling up to 90% of customer calls without a human agent.
Deepgram is the enterprise Voice AI platform for building and scaling real time voice applications on AWS. This product listing contains multiple versions of the nova-3 model which can each transcribe a set of languages. See version details for more information.
You will be billed $0.0077/min as described by https://deepgram.com/pricing. Private pricing available upon request.
Our APIs for Nova Speech to Text (STT) are natively available in the new SageMaker Bi-Directional Streaming API. Additional native touchpoints with Amazon Bedrock, Lex, and Amazon Connect make it simple to compose full voice experiences with the cloud services your teams already trust.
Customer reviews
Vikash K.
Speech-to-Text with Speaker Identification and PII Protection
Reviewed on Oct 06, 2026
Review provided by G2
What do you like best about the product?
As an AI engineer building an insurance claims platform, we needed a way for field adjusters to dictate property damage briefs directly from noisy job sites without manually typing on a mobile device. I really appreciate AssemblyAI's latest Universal-1 model, which is incredibly reliable against background noise like wind and traffic. It accurately captures complex alphanumeric policy codes and specific insurance terminology. The Python SDK and the asynchronous audio processing in our FastAPI backend are very helpful. Their speaker diarization feature is particularly valuable; it separates voices, making transcription of adjuster interviews with homeowners much easier for downstream processing. The PII redaction tool is also crucial for compliance, automatically masking sensitive information from recordings before they reach our database, which keeps our system secure.
What do you dislike about the product?
Implementing AssemblyAI's real-time streaming WebSocket API is notably more complex than using their standard asynchronous endpoint. Handling WebSocket connections, especially with mobile network drops, requires unexpected engineering overhead. It would help if their Python SDK offered more robust auto-reconnecting wrappers for unstable cellular connections. Additionally, their LeMUR framework, meant to apply LLMs directly to transcribed audio, doesn't fit our specialized needs. We already use our own customized LangChain orchestration to extract specific data and calculate exact Actual Cash Value (ACV) financial math from the transcripts. Because of our strict deterministic requirements, we bypass LeMUR entirely, making that portion of their platform redundant for our tech stack.
What problems is the product solving and how is that benefiting you?
AssemblyAI solves a massive data entry bottleneck for our mobile workforce. Before this, adjusters spent hours manually typing out add-on property damage assessments, which was tedious and sometimes physically unsafe when inspecting structurally compromised roofs. Now, they simply dictate a summary of the damage into their app, and AssemblyAI translates it flawlessly into text. This eliminates manual typing and drastically accelerates the speed at which our downstream pricing agents can analyze a claim.
Chaitanya T.
Veryi Useful Speech-to-Text Conversion for College Projects
Reviewed on Sep 26, 2026
Review provided by G2
What do you like best about the product?
I have used this for my college project where I found it very much useful for the speech to text conversion
What do you dislike about the product?
The downsides of this is the limited free tier credits,you may need to pay for it if you go out of free credit tiers
What problems is the product solving and how is that benefiting you?
Basically the speech to text conversion is a difficult task because you need to identify where the voice of a person is relevant as human voice consists of many unwanted sounds and fillers, which is hard to tackle.
Information Technology and Services
Easy to use
Reviewed on Sep 20, 2026
Review provided by G2
What do you like best about the product?
I really like how easy AssemblyAI is to use and how accurate and reliable its speech-to-text technology is. The APIs are well-designed, making it simple to integrate powerful AI features into applications.
What do you dislike about the product?
The pricing can feel a little high for smaller projects or teams with limited usage. It would also be helpful to have more flexibility in pricing plans for low-volume users.
What problems is the product solving and how is that benefiting you?
AssemblyAI makes it simpler to convert and analyze speech data by offering accurate, easy-to-integrate AI APIs. It saves development time and helps me add transcription and audio intelligence features without needing to build and maintain complex speech-processing systems from scratch.
Jeet S.
Easy-to-Test Voice Agents with a Great Playground and Real-Time Transcription
Reviewed on Sep 09, 2026
Review provided by G2
What do you like best about the product?
What I like most about AssemblyAI is how easy it is to use and test. I especially like the Playground because I can quickly try voice agents and adjust the prompt, voice, and other settings without needing much setup. The templates are also useful for trying different use cases.
I also appreciate that there’s solid developer support through the API, plus Python and JS options. The $50 credit is a nice bonus because it gives enough room to test the platform properly. Real-time transcription with interruption controls is another feature I personally like, and it’s something I didn’t see in others. To be honest Platform walkthrough was great , so i will say UI UX of the platform was self explanatory but overwhelming.
I also appreciate that there’s solid developer support through the API, plus Python and JS options. The $50 credit is a nice bonus because it gives enough room to test the platform properly. Real-time transcription with interruption controls is another feature I personally like, and it’s something I didn’t see in others. To be honest Platform walkthrough was great , so i will say UI UX of the platform was self explanatory but overwhelming.
What do you dislike about the product?
Some settings need a bit of testing before I can get the output I’m looking for. I also feel the pricing can become expensive as usage increases, especially on bigger projects.
What problems is the product solving and how is that benefiting you?
AssemblyAI solves my biggest problem: the trial-and-error that comes with live implementation. I can use the playground, and it helps me understand what I need and which metrics I can control. It also offers more detailed controls like Min Silence, Max Silence, Speaker diarization, Medical mode, Voice focus, Turn silence, VAD threshold, and Interruption delay. When I’m not fully aware of the workflow, it provides ready-made templates, which makes my life much easier.
Arpan s.
Fast, accurate speech-to-text API that was simple to set up in Python
Reviewed on Sep 04, 2026
Review provided by G2
What do you like best about the product?
The transcription accuracy is remarkably high, even when handling background noise and diverse speaker accents. The Python SDK makes integration simple and fast, requiring just a few lines of code to get running. Having built-in speech intelligence features like speaker diarization, auto-punctuation, and content summarisation directly within the API response saves substantial development and pipeline orchestration time.
What do you dislike about the product?
Real-time streaming transcription can occasionally encounter brief latency spikes during fluctuating network condition compared to asynchronous batch processing. in addition, the usage-based pricing can scale up quickly when running high volume production jobs across thousands of audio hours, and accuracy for uncommon regional dialects and technical domain jargon could be enhanced further.
What problems is the product solving and how is that benefiting you?
We had a ton of recorded call logs and audio meetings that were basically dead data because nobody had the time to sit and manually take notes. We thought about hosting our own whisper model on an EC2 instance, but dealing with GPU costs and pipeline maintenance wasn't the headache. Offloading it to AssemblyAI gave us a fast, hands-off pipeline. Now the audio gets transcribed, split by speaker, and pushed directly into our search index in just a few minutes without needing any dedicated server babysitting.