Voice AI Companies: The Leaders Across Models, Infrastructure and Agents

Voice artificial intelligence (AI) companies build software that holds spoken conversations, from the speech models underneath to agents that answer phone calls for clinics, restaurants and contact centers. Funding for the category climbed sharply in 2026, and the companies now sort into a few clear layers.

Below, we cover the leading voice AI companies by category, how to evaluate a vendor and where the next voice AI companies are likely to emerge.

What Is a Voice AI Company?

A voice AI company builds software that hears, understands and responds in speech. Vendors sell underlying speech-to-text (STT) and text-to-speech (TTS) models, real-time audio transport and orchestration or a voice-first product that completes a workflow. Agent companies sell finished agents that answer calls for clinics, restaurants or contact centers, often priced per conversation or resolution.

Live interactions use cascaded models or one speech-to-speech model. In a cascade, STT transcribes the caller, a large language model (LLM) decides what to say and a TTS model speaks. Streaming can finish the loop in under a second. Speech-to-speech preserves tone and hesitation, and it gives teams less control over tool calling and voice choice. Production systems must distinguish mid-sentence pauses from finished turns to avoid interrupting callers or leaving them waiting. Turn-taking is what separates agents from keypress-based interactive voice response (IVR) phone trees and legacy assistants built for timers and simple commands.

24 Notable Voice AI Companies by Category

Voice AI companies cluster by the layer of the stack they sell, and each layer prices differently. Models charge by character or audio hour, infrastructure companies by minute and agents often by conversation or resolution. Model companies increasingly ship full agents. Each category lists companies alphabetically, with funding figures only where primary sources confirm them.

5 Speech and Voice Models

Most companies at this layer now sell both recognition and synthesis, and the leaders are moving up the stack into full agents. The five companies below show that shift:

  1. AssemblyAI. Built its business on speech-to-text and lists a TTS product as coming soon.
  2. Cartesia. Sells its own STT and TTS models and packages them into Line, a complete voice agent product.
  3. Deepgram. Sells STT and TTS models plus a Voice Agent API, and raised a $130 million Series C at a $1.3 billion valuation in January 2026.
  4. ElevenLabs. Started with TTS and now sells full agents, backed by a $500 million Series D at an $11 billion valuation in February 2026 on annual recurring revenue (ARR) that passed $330 million at the end of 2025.
  5. Hume AI. Sold expressive TTS and speech-to-speech application programming interfaces (APIs) until its chief executive and several engineers joined Google DeepMind in January 2026, and it is shutting down those APIs on November 13, 2026.

Hume’s exit is the clearest sign that a model company needs more than a strong model to stay independent.

4 Voice Agent Infrastructure

Infrastructure companies move audio in real time, connect phone numbers and pass each turn of a call from STT to the LLM to TTS, and nearly all of them charge per minute. The four companies below cover the main options, from bundled pricing to open-source frameworks:

  1. Bland. Bundles models and phone lines into one per-minute price, which keeps budgeting simple.
  2. Daily. Maintains the open-source Pipecat framework and bills hosting separately from models.
  3. LiveKit. Runs an open-source real-time audio platform that powers OpenAI's ChatGPT voice mode, and reached a $1 billion valuation on a $100 million round in January 2026.
  4. Vapi. Charges a flat per-minute hosting fee, passes model costs through without markup and reached a $500 million valuation in May 2026 after Amazon Ring chose it over 40 rivals.

Bundled pricing is simpler to budget, and unbundled pricing lets teams swap models and see what each part of a call costs.

5 Customer Service Voice Agents

Customer service is the most crowded agent category in voice AI, and a growing number of its companies price on resolved conversations. The five companies below lead the category:

  1. Cresta. Sells AI agents and live agent assistance to contact centers, and reached $100 million in annual revenue in April 2026.
  2. Decagon. Sells customer service agents to enterprises including Block and Deutsche Telekom.
  3. Parloa. Sells voice agents to enterprise contact centers, with customers including Allianz.
  4. PolyAI. Builds voice agents for enterprise contact centers.
  5. Sierra. Charges per resolved conversation and raised a new round in May 2026.

Sierra's pricing shows where the category is heading, with agents billed against what a human rep would cost.

6 Vertical Voice Applications

Vertical companies own one phone workflow inside an industry's system of record. The six companies below cover healthcare, restaurants and collections:

  1. Assort Health. Handles patient intake, referrals, refills and eligibility inside athena and Epic, and raised a $120 million Series C in June 2026.
  2. ConverseNow. Runs phone and drive-thru ordering for restaurants.
  3. Hippocratic AI. Sells patient-facing agents to health systems.
  4. Prosper AI. Builds voice agents for healthcare providers' phone workflows, and raised a $30 million Series A in June 2026.
  5. Skit.ai. Handles collections calls, prices on performance and says it works within the Fair Debt Collection Practices Act (FDCPA) and the Telephone Consumer Protection Act (TCPA).
  6. Slang AI. Handles guest calls for restaurants.

Vertical agents are hard to replace once installed, because they run inside software the business already depends on.

4 Big Tech and Frontier Labs

Big tech companies and frontier labs collapse speech-to-speech pipelines into one API call and charge per audio token or minute. The four companies below sell voice directly to developers and enterprises:

  1. Google. Sells Gemini native audio through Vertex AI.
  2. Microsoft. Made Voice Live API generally available at Ignite 2025, and its Nuance unit sells Dragon Copilot for clinical use.
  3. OpenAI. Released GPT-Realtime-2 in May 2026.
  4. SoundHound AI. Moved deeper into contact centers by acquiring LivePerson, a deal that closed in September 2026.

Lab APIs are easy to adopt, and per-token costs get hard to forecast at contact-center volume.

How to Evaluate a Voice AI Company

People answer each other in a fraction of a second and judge agents by that habit. Whether you're buying or building, five checks cover performance at your target concurrency, with response times measured at the 95th percentile.

1. Latency and Turn-Taking

Response time runs from the caller's last word to the agent's first audio, and interruptions and mid-sentence pauses are the harder tests. Mistaking pauses for finished turns interrupts callers. Waiting too long feels like a hold queue.

2. Accuracy in Real Conditions

Public-leaderboard word error rate (WER) reflects that leaderboard's audio and varies with accents and noise. Your own recordings test WER against your callers' accents, background noise and domain vocabulary.

3. All-In Cost per Minute

Vendors either bundle the LLM, speech models and phone line or add hosting fees to models passed through at cost. Fair comparisons price your own calls, including silence and hold time.

4. Compliance and Consent

TCPA rules cover AI-generated artificial or prerecorded voices, so outbound calls need prior express consent. Healthcare deployments need a business associate agreement (BAA) under the Health Insurance Portability and Accountability Act (HIPAA). Article 50 of the European Union (EU) AI Act requires telling people they're talking to AI and disclosing deepfakes.

5. Workflow Ownership

A product that completes the task inside the customer's systems, by booking appointments in Epic or posting collections-ledger payments, bills on outcomes and stays installed. Agents that only talk before handing off to humans compete on price per minute.

Why Voice AI Funding Is Accelerating

Voice AI startups raised about $7 billion in the first quarter of 2026, up from roughly $1 billion a year earlier. Investors are betting on labor budgets. The United States employs about 2.67 million customer service representatives at a median wage near $45,000, with a projected five percent employment decline over the next decade as automation absorbs routine volume. Customers still choose phone calls for complex problems, and voice-agent revenue tracks minutes or resolutions handled.

Customer service leaders' AI spending grew faster than their overall budgets. Most agent-company ARR figures are self-reported, so any revenue multiple built on them is directional at best.

Where the Next Voice AI Companies Will Emerge

Model companies are moving into agents, frontier labs into orchestration and enterprise agent companies toward the largest contact centers. Those moves leave regulated edges and non-English markets less crowded. Founders are starting companies in five openings.

  • Stack consolidation. Cartesia's Line, the Deepgram Voice Agent API and ElevenLabs Agents sell complete agents on their own models, so thin orchestration startups now compete with their suppliers.
  • Native speech-to-speech. Models that listen and speak at once remove STT and TTS vendors from simple calls. Cascaded pipelines still dominate enterprise deployments because teams can inspect each stage
  • Regulated, call-heavy verticals. Insurers use voice agents for first notice of loss, cities are moving nonemergency 311 calls to AI and healthcare intake keeps drawing seed and Series A rounds.
  • Non-English and multilingual markets. Ringg and Sarvam have raised in India, and models already support dozens of languages for local telephony, regulation and dialect.
  • Evaluation, monitoring and deepfake detection. Coval and Hamming sell voice-agent testing, and Modulate and Reality Defender sell deepfake detection for agents that take payment and authentication calls.

Each opening sits where the funded leaders' current products stop, whether that is below the largest contact centers, outside English or in the checks that run after a call ends.

What the Voice AI Market Means for Founders

Models are commoditizing fastest. Full agents from model companies, Hume's team moving to Google DeepMind and frontier labs' single-model pipelines keep pushing down raw-speech prices. Workflow agents hold up better. Agents installed in Epic, collections ledgers or restaurant order systems are hard to replace, because leaving means rebuilding the integration.

We watch whether native speech-to-speech squeezes thin orchestration layers and whether agent-company ARR holds up under audited disclosures. If you're an early stage founder looking for seed or Series A support for a workflow-owning voice agent, reach out to us to see if we'd be a good fit.

Frequently Asked Questions

What is the biggest voice AI company?

ElevenLabs is the most valuable independent voice AI company after a September 2026 employee tender valued it at $22 billion. SoundHound AI is publicly traded and reports quarterly revenue, so its figures don't line up with private companies' ARR.

How do voice AI companies make money?

Model companies charge per character of synthesized speech or per hour of transcribed audio, infrastructure companies charge per minute of live call time and agent companies increasingly charge per resolved conversation or outcome. ElevenLabs sells TTS per thousand characters and its agents per minute, Sierra charges for resolved outcomes and Skit.ai prices collections on performance. Pricing shows which budget a company sells into. Per-character and per-minute products compete with other software, and per-resolution products compete with a human agent's hourly cost.

What is the difference between voice AI and conversational AI?

Conversational AI holds human-like conversations across text, voice or both, and voice AI is the spoken-language part of it. Voice systems must transcribe, decide and speak within about a second and handle interruptions and background noise along the way, constraints text chatbots don't face.

Can voice AI replace call center agents?

Voice AI is absorbing routine call volume faster than it replaces whole jobs. Customer service jobs face a projected five percent decline by 2035 as automation absorbs routine contacts. Klarna's AI chat assistant reportedly did the work of hundreds of agents as the company shrank its workforce, then resumed hiring humans amid concerns that cost had outweighed quality.

Congrats to Lotus AI and Outtake on Making Forbes Next Billion-Dollar Startups List

CRV proudly co-led Lotus AI’s Series A and our firm led Outtake’s Series A and joined the board in February 2025. We also backed Outtake during its Series B, so we’re thrilled to see both teams make this year’s list.” to “CRV proudly co-led Lotus AI’s Series A and our firm led Outtake’s Series A, joined the board and backed Outtake during its Series B, so we’re thrilled to see both teams make this year’s list.

CRV invests in founding teams at the beginning of their journeys, leading Seed and Series A rounds in amazing companies. We’ve backed more than 750 companies early on including DoorDash (another Next-Billion alum), Mercury and Vercel.

Congrats to both Lotus AI and Outtake on being named to Forbes’ Next-Billion Dollar Startups list.

Cookie Preferences

Your Privacy Matters to Us

We use cookies and similar technologies on this site, employed by CRV and our partners, to support core features and help us understand how visitors engage with our content. For details, please review our Privacy Policy