Voice AI companies fall into five layers: makers of speech and voice models, platforms for building voice agents, enterprise customer-service platforms, open-source and phone-line building blocks, and specialists that build agents for one market. Single rounds in 2026 ran to hundreds of millions of dollars. In Africa, a few companies, including ShiftMate, focus on local accents.


“We're ShiftMate, a voice agent company based in Umhlanga, South Africa, and one of the African names on this map. This guide sets out who's who in voice AI by layer, from AI labs to platforms to specialists, with funding and deals, and where we think the companies that last will come from.”
Voice AI companies do one of five jobs, and a company's layer tells you what you'd be buying: a part, a toolkit, a managed service or a finished result. Every company fact here was checked on 16 November 2026 against the company's own page or a named news outlet, with the link beside it.
| Layer | What they sell | How they usually charge | Examples |
|---|---|---|---|
| Speech and voice models | Speech recognition, text-to-speech, speech-to-speech, language tools | Per minute of audio, per token or per plan, where published | ElevenLabs, Deepgram, Rime, the big AI labs' speech APIs; Lelapa AI and Intron for African languages and accents |
| Agent platforms | Tools to build, run and monitor voice agents | Per minute, often with parts billed on top | Retell AI, Vapi, Bland AI, ElevenLabs Agents, Synthflow |
| Enterprise customer-service platforms | Voice and chat agents for large contact centres, with services | Annual contract, or per outcome | PolyAI, Sierra, Parloa, NiCE Cognigy, Decagon, Wonderful |
| Building blocks | Open-source frameworks, real-time media, phone lines | Per minute of hosting or call time | LiveKit, Pipecat (Daily), Twilio |
| Market specialists | Agents built, tested and run for one market or language group | Per business or per project | ShiftMate and Untapped AI (South Africa) |
A platform gives you tools: you still write the script, connect your systems, test the agent and watch its calls. An enterprise platform does more of that, at contract prices. A specialist sells the result. Most lists of voice AI companies leave out two jobs that decide whether an agent works: testing it on difficult calls before customers hear it, and training the people who take over from it. That's where we work. Weighing a build? See how to build an AI voice agent.
The voice AI companies a business is most likely to meet when it shops for a voice agent are the ten platforms below, plus the specialists further down. Facts checked 16 November 2026; speed and language figures are each company's claim.
A newer South African voice AI company, based in Umhlanga, building for where the market is going: voice AI that is judged on what it produces, not on how many minutes it talks. What it offers: an AI outbound sales agent that calls, qualifies and books your leads (priced in rand), inbound voice agents built for each business, and AI practice calls that train and test call centre and sales teams. Why it's different: built for South African callers and rules (POPIA, the national opt-out registry), natural South African voices, AI sales agents that must pass difficult practice calls before they phone anyone, and one engine that sells, trains and interviews. Who it's for: call centres and sales teams that want their people trained, their leads worked and the low-value calls taken off their operators, and businesses that could never afford a call centre. If you want to build agents yourself, the developer platforms below are built for that.
We list ourselves first because it's our page. It isn't a ranking: the platforms below follow in no particular order.
A voice agent platform for contact centres: a no-code studio plus APIs. $0.07 to $0.31 a minute, itemised by part (pricing); claims replies in about 600 milliseconds (its site). Says it passed $100 million in annualised revenue on one $4.6 million seed round (source).
A developer platform for building and running voice agents: $0.05 a minute for hosting, with speech, AI model and voice passed through at cost (pricing). TechCrunch reported a valuation of about $500 million, citing a person familiar (TechCrunch).
An enterprise phone-agent platform on its own hosted models; says it speaks 40+ languages natively (its site). $0.14 a minute, or $0.12 plus $299 a month, phone lines extra (pricing).
The voice company's agent platform, listing 90+ languages (its site). Plans from free to $990 a month, then $0.08 a minute; the AI model is billed separately (pricing).
Voice agents for contact centres, BPOs and multi-location brands, with white-label workspaces. Annual enterprise contracts only: no free plan, no per-minute rate (pricing).
London-based enterprise voice agents for customer service, on its own models. Custom pricing; total funding over $200 million (source).
Enterprise customer-experience agents across voice, chat, SMS, WhatsApp and email, priced on outcomes: "pay for a job well done" (Sierra).
A Berlin platform to "build, test, and scale voice and chat AI agents" for enterprise customer experience. Custom pricing (Parloa).
Voice and chat agents for enterprise contact centres, with low-code design. Custom pricing; owned by NiCE since September 2025 (source).
One API combining speech recognition, an AI model and a voice, at $0.075 a minute (Standard) (pricing).
To choose between them, see the best AI voice agents compared.
Behind most voice agents sit three kinds of supplier: speech and voice model makers, open-source frameworks with real-time hosting, and phone networks. Their prices and outages reach you through your platform.
Voice models and an agent platform. Raised a $500 million Series D at an $11 billion valuation in February 2026, led by Sequoia (source). A September 2026 employee share sale valued it at $22 billion (not new money) (Reuters).
Speech recognition and a voice agent API. It says its turn detection spots the end of a caller's turn in about 260 milliseconds (its claim).
Voices for enterprise customer calls. Raised a $24 million Series A in July 2026, led by M13 with Twilio Ventures taking part (TechCrunch).
An open-source agents framework and media server; LiveKit Cloud charges $0.01 per agent-session minute (pricing). Raised a $100 million Series C at a $1 billion valuation in January 2026, led by Index Ventures (source).
Pipecat is an open-source voice agent framework maintained by Daily (Daily). Since Deepgram bought Daily, it stays open source under an independent foundation (The Next Web).
The big AI labs also sell speech-to-speech models directly, priced per token, with phone lines and hosting left to you (OpenAI's API pricing). See AI voice agent pricing for what each part costs.
The largest voice AI funding rounds we could verify for 2025 and 2026 went to enterprise customer-service agents and voice model makers: Sierra ($950 million), Wonderful ($550 million) and ElevenLabs ($500 million) lead the table. CB Insights counted $2.1 billion raised by voice AI companies in 2024, up from $264 million in 2023, and nearly $500 million in the first quarter of 2025 (CB Insights).
| Company | What it does | Round | When | Led by |
|---|---|---|---|---|
| Sierra | Customer-service agents, chat and voice | $950 million, valued above $15 billion (source) | May 2026 | Tiger Global and GV |
| Wonderful | Enterprise AI agents, began with voice support | $550 million Series C, $5 billion valuation (source) | September 2026 | Insight Partners |
| ElevenLabs | Voice models and agents | $500 million Series D, $11 billion valuation (source) | February 2026 | Sequoia Capital |
| Parloa | Customer-service agents | $350 million Series D, $3 billion valuation (source) | January 2026 | General Catalyst |
| Decagon | Customer-service agents | $250 million Series D, $4.5 billion valuation (source) | January 2026 | Coatue and Index Ventures |
| ElevenLabs | Voice models and agents | $180 million Series C, about $3.3 billion valuation (Crunchbase News) | January 2025 | ICONIQ and a16z |
| Deepgram | Speech recognition, voice agent API | $130 million Series C, $1.3 billion valuation (source) | January 2026 | AVP |
| Parloa | Customer-service agents | $120 million Series C, $1 billion valuation (Observer) | May 2025 | General Catalyst |
| LiveKit | Real-time voice infrastructure | $100 million Series C, $1 billion valuation (source) | January 2026 | Index Ventures |
| PolyAI | Enterprise voice agents | $86 million Series D (source) | December 2025 | Georgian, Hedosophia, Khosla Ventures |
| Bland AI | Enterprise phone agents | $50 million Series C (source) | June 2026 | Dell Technologies Capital |
| Vapi | Developer voice agent platform | $50 million Series B (source) | May 2026 | Peak XV |
| Rime | Voices for customer calls | $24 million Series A (TechCrunch) | July 2026 | M13 |
| Gnani.ai | Multilingual voice agents, India | $10 million Series B (Entrackr) | March 2026 | Aavishkaar Capital |
Among these voice AI funding rounds, the biggest cheques go to companies selling a whole customer-service result to enterprises, and revenue doesn't track funding, as Retell AI's $4.6 million seed shows (source). We don't quote a single total for 2026: the trackers define "voice AI" differently, and their totals disagree.
At least eleven voice and conversational AI acquisitions were announced between July 2025 and November 2026, several of them, including NiCE, Salesforce and Meta, large companies buying a voice layer they didn't have. Prices only where published.
| Buyer | Bought | When | Price |
|---|---|---|---|
| Meta | PlayAI (voice generation) | July 2025 | Undisclosed; PlayAI had raised $23.7 million (Crunchbase News) |
| NiCE | Cognigy (conversational and voice AI) | Closed September 2025 | $955 million (Cognigy) |
| Hippocratic AI | Grove AI (patient-recruitment agent) | January 2026 | Undisclosed (Fierce Healthcare) |
| SoundHound AI | LivePerson | Announced April 2026 | $43 million equity value, $250 million enterprise value (SoundHound) |
| Salesforce | Fin (formerly Intercom) | Announced June 2026 | Terms not checked (Salesforce) |
| Nurix AI | Verloop.io (voice and chat agents, India) | July 2026 | Undisclosed (Inc42) |
| LiveKit | Loophole Labs (agent infrastructure) | September 2026 | Undisclosed (press release) |
| Automation Anywhere | Boost.ai | Agreed October 2026 | Undisclosed (Unite.AI) |
| Deepgram | Daily, maker of Pipecat | October 2026 | Undisclosed (TechCrunch) |
| Salesforce | Cresta (voice-first customer experience) | Agreed November 2026 | Not disclosed (Salesforce) |
| Worker.ai | DeltaGen (real-time voice AI, Prague) | November 2026 | Undisclosed (The Recursive) |
The pattern: contact-centre and CRM software companies are buying voice to sell inside products you may already use. GV's Tom Hulme told Crunchbase News that voice AI startups are "ripe for acquisition" (Crunchbase News). For a buyer: the company you sign with today may belong to someone else next year, so keep your numbers, scripts and recordings exportable.
The VCs investing in voice AI most visibly are Andreessen Horowitz (a16z), Sequoia Capital, General Catalyst, Index Ventures and Insight Partners, with corporate investors such as Salesforce, Nvidia and Twilio Ventures alongside them. Each is backed by a round above or a named source, checked 16 November 2026.
| Investor | Voice AI bets on record |
|---|---|
| Andreessen Horowitz (a16z) | Co-led ElevenLabs' Series C and added in its Series D; backs Decagon; publishes the clearest thesis on voice agents (a16z) |
| Sequoia Capital | Led ElevenLabs' $500 million Series D; backs Sierra |
| General Catalyst | Led Parloa's Series C (May 2025) and Series D (January 2026) |
| Index Ventures | Co-led Decagon's Series D; led LiveKit's Series C; existing investor in Wonderful |
| Insight Partners | Led Wonderful's $550 million Series C |
| Corporate investors | Salesforce (ElevenLabs, Wonderful), Nvidia (ElevenLabs), Twilio Ventures (Rime), Dell Technologies Capital (Bland AI) |
a16z's 2025 update says "the most natural early categories for voice agents typically have high existing call center/BPO spend", and that realistic voice agents can act as "simulators" for training (a16z). Both describe South Africa's call centre industry well.
Yes. A small group of voice AI companies in South Africa and Nigeria focus on African accents and languages, because speech systems trained mostly on American and European speech struggle here: off-the-shelf speech recognition makes several times more errors on South African-accented English than on standard US benchmarks (AfriSpeech-200, 2023).
Umhlanga, KwaZulu-Natal. An AI outbound sales agent that qualifies and books leads, inbound voice agents built for each business, and AI practice calls for call centre and sales teams.
Cape Town. Builds South African voice agents: it recorded 20 South African voices in a studio, works in English and Afrikaans with isiZulu and isiXhosa in development, and routes calls through South African data centres. Its CEO says drop-off is "significantly higher" with robotic or foreign accents (iAfrica).
Johannesburg. Builds speech and language tools for African languages, aimed at uses such as contact centres; raised a $2.5 million seed round in 2023 (MIT Technology Review).
Speech recognition for African accents; raised a $1.6 million pre-seed round in 2025 (Semafor).
Local firms also sell voice agents, mostly AI receptionists. We list what their sites say, not how well they work; rand prices for several are on WhichVoIP's roundup and on our pricing page.
| Company | What its site says | Languages it lists |
|---|---|---|
| Agent Help | AI receptionist, free one-week trial | English, Afrikaans, isiZulu |
| AiSky | AI voice reception for South African businesses | Claims all 11 official languages (per WhichVoIP) |
| OnDial | South African page; "POPIA-aligned" | English, isiZulu, isiXhosa, Afrikaans |
| Telecloud | AI call centre on its own South African voice network | Afrikaans, isiZulu |
The National reported that Nutun's AI collections agent, Zoey, deals with more than 100,000 people a day, and that Concentrix stretched training from three weeks to eight because of AI (BPESA). People then take the harder calls, which is why we put AI practice calls beside our agents.
“The market responds better to natural, authentic South African voices. An American or British voice on a South African line sounds like a scam call. So our sales agents always sound South African, from the first word to the last, even on UK or US campaigns.”
Choose a voice AI company by matching its layer to your job, then testing it on your own calls. Use this checklist with anyone, us included.
Funding isn't quality. Language counts aren't accents: "90+ languages" tells you nothing about a caller from Chatsworth switching between English and isiZulu. And your compliance is yours, not the supplier's. South Africa's national opt-out registry is being phased in: direct marketers register with the National Consumer Commission from 15 September 2026, cleanse their lists monthly (free cleansing December 2026 to April 2027), and consumers can block marketers from May 2027 (NCC). This is general information, not legal advice: take legal advice for your own calls.
“We learned to measure the real answer, not the filler. A quick "Sure" makes a speed number look good while the caller is still waiting for what they asked. And we measure the round trip from a South African caller, not from a server next door to the AI.”
How agents break with real customers, and how to catch it first: why AI voice agents fail.
ShiftMate is a South African voice AI company that builds AI outbound sales agents, inbound voice agents made for each business, and AI practice calls that train call centre and sales teams, all on one engine.
We think the voice AI companies that last won't be those with the biggest rounds, but those whose agents keep working after the demo, on real callers. Among the five layers above, we're a market specialist: our AI outbound sales agent and AI practice calls are built and tested on South African calls.
Choosing between layers? Send us your calls and we'll say which layer fits, even when it isn't ours.
“Voice AI isn't replacing call centres. What it's doing is making voice available and affordable to businesses that could never afford a call centre, and supporting the call centres that exist: training their people, generating leads, and making the low-value calls their operators don't want to make. This space is changing and opening up.”
A voice AI company makes technology that understands or produces speech, or uses it to run phone conversations. Some make speech models, some sell platforms or enterprise agents, and some build agents for one market.
Among privately funded voice AI companies, by recent valuation, Sierra (above $15 billion, May 2026) and ElevenLabs ($11 billion at its February 2026 round; $22 billion in a September 2026 employee share sale) are among the largest. Wonderful ($5 billion), Decagon ($4.5 billion) and Parloa ($3 billion) follow. Checked 16 November 2026.
The largest priced deal was NiCE buying Cognigy for $955 million, closed in September 2025. SoundHound AI agreed to buy LivePerson at a $250 million enterprise value in April 2026. Salesforce agreed to buy Cresta in November 2026 without publishing a price, and Deepgram bought Daily, maker of Pipecat, in October 2026.
Andreessen Horowitz, Sequoia Capital, General Catalyst, Index Ventures and Insight Partners led many of the largest voice AI rounds of 2025 and 2026. Tiger Global, GV, Peak XV and Dell Technologies Capital also led rounds; Salesforce, Nvidia and Twilio Ventures invest as corporates.
Yes. ShiftMate builds voice agents and AI practice calls for South African businesses; Untapped AI builds South African voice agents in English and Afrikaans; Lelapa AI builds speech tools for African languages. Local firms such as Agent Help and Telecloud sell AI receptionists.
Published per-minute prices depend on what they include: Retell AI lists $0.07 to $0.31 a minute, built from itemised parts (voice engine, voice, AI model and a US$0.015 phone line; the $0.07 floor has no AI model or phone line), Vapi charges $0.05 for hosting alone, and LiveKit Cloud $0.01. Enterprise platforms sell contracts or charge per outcome. South African sellers list R849 to R7,500 a month (WhichVoIP, third-party). Checked 16 November 2026.
The common mistakes are choosing on funding or brand instead of a test on your own calls, comparing per-minute prices that cover different parts, trusting speed claims measured near the supplier's servers, assuming a language count means your callers' accents will work, and having no exit plan for numbers, scripts and recordings if the company is sold.
Yes, within the rules. POPIA section 69 needs consent for direct marketing by electronic communication, including automatic calling machines, with a narrow existing-customer exception. The national opt-out registry is being phased in: marketers register from 15 September 2026 and consumers can block them from May 2027. This is general information, so take legal advice.
Built in Umhlanga and tested on South African calls: sales agents, inbound agents made for each business, and practice calls for the people beside them.