AssemblyAI

AssemblyAI is a company focused on voice AI infrastructure, developing its own speech-to-text foundation models and expanding into a full voice agent technology stack.

AssemblyAI-TTS-1 claims to offer unlimited concurrency, meaning the server does not limit the number of simultaneous synthesis requests, and integrators need not pay extra for capacity planning or apply for quotas.

AssemblyAI-TTS-1's streaming response first-packet latency is as low as 90 milliseconds.

AssemblyAI-TTS-1 uses per-character billing at $15 per 1 million characters and does not require an annual contract.

The source material does not state whether AssemblyAI's ASR and TTS services have bundled pricing or joint optimization.

AssemblyAI's main business is on the speech recognition (ASR) side.

AssemblyAI-TTS-1 can be used without signing a contract and with unlimited concurrency.

AssemblyAI provides an official Node SDK that supports uploading audio files to obtain transcript text and webhook callbacks; Deepgram's Node SDK supports real-time streaming transcription; Google Cloud's Node client is complex to configure and requires a service account key and project settings.

Deepgram's prepaid package costs about $0.0043 per minute, and AssemblyAI offers a free quota of 100 minutes per month.

In a comparative evaluation, AssemblyAI leads in accuracy and API flexibility, while Otter.ai and Microsoft Teams have greater advantages in ease of use and office ecosystem integration.

AssemblyAI supports streaming real-time transcription via WebSocket and provides a Node.js example.

The typical pipeline involves uploading a meeting recording to the cloud, which automatically triggers a transcription service (such as AssemblyAI, Whisper API, or Otter.ai), passing the transcribed text to an LLM for action item extraction, and then writing the results into a task tool via API.

Voice data contains the speaker's voiceprint characteristics, which constitute biometric information; most mainstream ASR solutions rely on cloud-based APIs (such as Google Speech-to-Text and AssemblyAI), causing audio or text to be processed by third-party servers.

Off-the-shelf customer service bots lack access to system prompts, tool-calling logic, and RAG retrieval infrastructure, causing iteration speed to be constrained by the vendor.

The off-the-shelf customer service bot AssemblyAI initially purchased resolved only about 10% of conversations.

Joey is AssemblyAI's self-developed AI customer service agent, built on the Claude Agent SDK, capable of managing infrastructure, owning a file system, writing and debugging code, and calling tools.

Any customer who contacts AssemblyAI—whether through the website chat window, by contacting sales, or by emailing the support inbox—is first received by Joey rather than a human.

The team added voice capability to Joey using AssemblyAI's own Voice Agent API, chaining speech-to-text, LLM, and text-to-speech together via a single WebSocket connection.

Signing a BAA (Business Associate Agreement) automatically opts the account out of model training.

In the first week after Joey's deployment, the end-to-end resolution rate jumped from 10% to 80%, with a total cost of approximately $700 per month (tokens plus infrastructure).