AssemblyAI vs Krisp: pricing, features and which to choose

AssemblyAI was founded in late 2017 by machine-learning engineer Dylan Fox after he found existing speech models inaccurate and difficult for developers to use. Its cloud platform serves product and engineering teams through APIs for recorded transcription, real-time speech, speech understanding, guardrails, voice agents, and language-model access. In December 2023, TechCrunch reported that AssemblyAI had raised $50 million in an Accel-led round, bringing total funding to $115 million. The company said its paying customer base had grown 200% year over year to 4,000 brands, while more than 200,000 developers used its platform. AssemblyAI also reported handling roughly 25 million API calls and more than 10 terabytes of data daily.

AssemblyAI’s workflow begins with choosing a product and model, then sending recorded audio through an HTTP request or opening a WebSocket for live transcription. For asynchronous files, developers select Universal-2 at $0.15 per audio hour or Universal-3.5 Pro at $0.21, add options such as speaker diarization, entity detection, or Medical Mode, and receive a transcript after processing. Streaming applications open a session, send audio frames, consume partial and final results, and send a termination message when finished because idle WebSocket time remains billable. Voice applications can connect to the Voice Agent API, provide a system prompt and tools, exchange audio in both directions, and handle turn detection and interruptions. Teams can then pass transcripts to Speech Understanding or the LLM Gateway for structured analysis.

AssemblyAI’s main limitations are cost complexity and developer overhead. Unlike Deepgram and Speechmatics, it bills streaming by open WebSocket duration, so idle time counts; add-ons, multichannel audio, and separate LLM Gateway token charges can also complicate forecasts. Google Cloud Speech-to-Text and AWS Transcribe may suit organizations already committed to those clouds, while AssemblyAI is a standalone API platform rather than a broader cloud suite. Its terms permit use of customer data to maintain, develop, and improve services, including model training, subject to applicable opt-out arrangements, so privacy reviews matter for sensitive workloads. AssemblyAI offers $50 in eligible credits, no monthly minimums, EU speech endpoints at US prices, and a $4.50-per-hour Voice Agent API. Choose it for API-first voice products needing transcription plus speech intelligence; choose competitors for cloud consolidation or simpler procurement.

Krisp Technologies, Inc. develops software for real-time voice communication. Founded in 2017, the company says its technology is deployed on more than 200 million devices and processes more than 80 billion minutes of voice conversations monthly. Its product line extends beyond desktop noise filtering: Meeting AI handles transcription and notes, Contact Center AI supports agent productivity, and developer products provide voice models and real-time translation. Krisp raised a $5 million Series A in 2019, according to its year-in-review announcement. Headquartered in Berkeley, California, the company serves consumers, businesses, contact centers, and developers. This range means prospective customers should identify their use case before comparing plans, because the products have separate purposes and purchasing paths.

A typical Meeting AI workflow places Krisp between the user and Zoom, Teams, Google Meet, or another calling application as a virtual microphone and speaker. Noise cancellation runs locally. The optional AI assistant captures meeting audio, creates a transcript, and produces notes, action items, and searchable chat. Paid Meeting plans include bot-free recording and notes for in-person meetings. Integrations and webhooks can connect outputs with other workflows, while mobile access extends note-taking beyond a desktop call. Contact Center AI adds after-call summaries, monitoring, agent assist, speech analytics, and accent conversion for live customer conversations. Teams can use Krisp to improve what participants hear or reduce the manual work required after conversations end.

Krisp’s strongest distinction is its combination of local noise removal, bot-free meeting capture, accent conversion, and voice-agent infrastructure. Its cloud meeting features still require trust in a service handling transcripts and recordings, while Core storage is limited to 10 GB. The product line may also be confusing because Meeting AI, Contact Center AI, and developer APIs have separate plans and billing. Otter.ai and Fireflies.ai provide established meeting-transcript alternatives, while Zoom AI Companion may cost less for organizations already paying for Zoom. Krisp should be chosen by teams that need audio processing as well as meeting notes, especially contact centers and voice-agent builders. Teams seeking only basic transcription may find a narrower meeting tool easier to evaluate and administer.