Voice & Audio AI Market Map: 88 Players — From ElevenLabs, OpenAI & Suno to Supertone, Perso.ai & Qwen3-TTS

After text-generation LLMs reached mainstream adoption, the fastest-growing frontier for enterprise monetization and paid creator subscriptions has been Voice & Audio AI. Just two years ago, voice AI was largely confined to robotic text-to-speech (TTS) readers. Today, it has bifurcated into four distinct multi-billion-dollar verticals: zero-shot voice cloning that replicates breath and emotional prosody from a 3-second audio clip, real-time conversational phone agents that handle interruptions (barge-in) in under 300ms, end-to-end AI music studios generating vocals, accompaniment, and mastered stems from a single prompt, and automated video dubbing that synchronizes lip movements (lip-sync) across dozens of languages.

Today we are publishing the Voice & Audio AI Market Map, a comprehensive directory of 88 core players across 5 regions (US/Global, KR, JP, CN, EU), structured into 4 primary categories—① TTS & Voice Cloning, ② Real-Time Voice Agents & STT, ③ AI Music & Sound Effects, and ④ Video Dubbing & Localization—and 4 tiers (Tier 1, Tier 2, Rising, Open Source). Every entry includes a 3-tier verification badge (✔ Official/Verified, △ Estimated/Reported, ✘ Unconfirmed) and primary source URLs (src).

• Explore the Interactive Market Map: Voice & Audio AI Market Map (/en/maps/voice-audio-ai/)

---

1. Market Architecture: Why Capital Is Flooding Into Voice & Audio AI

Revenue in the audio AI ecosystem flows through two primary engines: creator and media production SaaS (tiered monthly subscriptions) and real-time telephony, conversational agent, and infrastructure APIs (usage-based billing per minute or per character).

1. Unicorn Dominance vs. Open-Weight Disruption in TTS & Voice Cloning: US/UK-based ElevenLabs reached a $3.3B valuation in its Series C while surpassing $100M+ in ARR. Yet a wave of open-weight models capable of delivering 90%+ of commercial quality on a single local GPU—including Qwen3-TTS (Alibaba), VoxCPM2 (OpenBMB), GPT-SoVITS, CosyVoice 2, Fish Speech (Fish Audio), and Chatterbox (Resemble AI)—is driving the unit cost of bulk audiobook narration and game NPC dialogue down by more than 10x. 2. The "Sub-500ms" Latency War in Real-Time Voice Agents: Humans perceive conversational turn-taking as natural only when round-trip latency stays under 300–500ms. Legacy chained pipelines (STT → LLM → TTS) suffered from 1.5–3.0 second delays. Now, combining ultra-fast engines like Cartesia (Sonic · 40–75ms TTFA), Deepgram (Nova-3 + Aura), and OpenAI Realtime API (native Speech-to-Speech) with telephony/WebRTC orchestration layers—Retell AI, Vapi, Bland AI, LiveKit Agents, and Pipecat (Daily)—has turned AI call centers, clinic scheduling, and logistics dispatch into one of the fastest-growing B2B revenue streams in AI. 3. AI Music Explosion, Major Label Lawsuits & "License-Clean" Realignment: Suno and Udio have scaled to tens of millions of monthly generated tracks, but face landmark copyright infringement lawsuits from the RIAA (Universal, Sony, and Warner Music). Consequently, enterprise buyers in advertising, gaming, and broadcast BGM are gravitating toward models trained on licensed catalogs—such as Stability Audio (Stable Audio 2.0), Soundraw, Korea's Pozalabs and Neutune (MixAudio), and Kunlun Tech's Mureka (SkyMusic). 4. Video Dubbing & Localization Fueled by K-Content, Creators & Cross-Border Commerce: Since YouTube rolled out multi-language audio tracks globally, demand for automated dubbing has surged. Global leaders HeyGen (Video Translate), Rask AI, Sync Labs, ElevenLabs Dubbing Studio, and Camb.ai compete alongside Korean powerhouses ESTsoft (Perso.ai), Supertone (HYBE), Typecast (Neosapience), Naver (Papago Dubbing · CLOVA Dubbing), Hudson AI (acting-based dubbing), Voithru (Jamake), and Twigfarm (HeyJoy).

---

2. Category Leaders: Per-Minute Pricing & Latency Benchmarks

Below is a head-to-head comparison of flagship players featured in the Voice & Audio AI Market Map:

| Category | Flagship Player (Region) | Key Metric / Valuation (Verif.) | Latency (TTFA / E2E) | Representative Pricing & Billing Formula | |---|---|---|---|---| | ① TTS & Voice Cloning | ElevenLabs (US/Global) | $3.3B Valuation · $100M+ ARR (✔) | Flash v2.5 ~75ms | $5–$330/mo · ~$0.06–$0.18/min | | ① TTS & Voice Cloning | Cartesia (US/Global) | $91M Total Funding · SSM Architecture (✔) | Sonic TTFA ~40ms | $5–$299/mo · Low-latency streaming credits | | ① TTS & Voice Cloning | Supertone (KR) | Acquired by HYBE for KRW 45B (✔) | Real-time voice changer ~45ms | Supertone Shift · Play SaaS & B2B IP licensing | | ① TTS & Voice Cloning | Typecast (KR) | 2M+ Global Users (✔) | Studio synthesis 1–2s | $8.99–$89/mo · Emotion & character acting TTS | | ① TTS & Voice Cloning | Qwen3-TTS / CosyVoice (CN · OS) | Alibaba Open-Source (✔) | Streaming ~150ms | Apache 2.0 self-hosted free (GPU compute only) | | ② Voice Agents & STT | Deepgram (US/Global) | $85M+ Raised · Nova-3 (✔) | STT <200ms · Aura TTS <150ms | STT $0.0043/min · TTS $0.015/1K chars | | ② Voice Agents & STT | Retell AI / Vapi / Bland (US/Global) | Tens of millions of monthly calls (△) | E2E turn-taking 500–700ms | $0.05–$0.15/min all-in (STT+LLM+TTS+Telephony) | | ② Voice Agents & STT | LiveKit Agents (US/Global · OS) | $45M Series B · OpenAI Realtime Infra (✔) | WebRTC edge transport <100ms | Open-source framework free + Cloud per-minute | | ② Voice Agents & STT | Return Zero (KR) | Top Korean STT (VITO · Callabo) (✔) | Real-time Korean call/meeting STT | B2C freemium + enterprise AICC deployment | | ③ AI Music & SFX | Suno / Udio (US/Global) | Suno $125M Raised · $500M Val. (✔) | Full song in 15–30s | $10–$30/mo (commercial rights on paid plans) | | ③ AI Music & SFX | Stability Audio (US/Global) | Stable Audio 2.0 / Open (✔) | 3-min stereo track in ~20s | $12–$99/mo · Trained on licensed AudioSparx data | | ③ AI Music & SFX | Pozalabs / Neutune (KR) | Leading Korean B2B & Creator Music AI (✔) | Instant stem & track generation | LAIVE · MixAudio subscriptions & B2B media sync | | ④ Video Dubbing | HeyGen / Rask AI (US/Global) | HeyGen $500M Val. · $35M+ ARR (✔) | Lip-synced video dubbing in minutes | ~$1–$3/min credit deduction · Team/Enterprise plans | | ④ Video Dubbing | ESTsoft Perso.ai (KR) | 200K+ Global Users · K-Dubbing Leader (✔) | Multi-speaker lip-sync automation | Monthly subscription + per-minute credits |

---

3. The 5-Layer Cost Stack of a 1-Minute AI Voice Agent Call

A common pitfall when budgeting an AI call center or automated phone agent is equating the headline platform fee with total cost. In production, five distinct layers are billed concurrently for every minute of conversation:

• Layer 1 — Audio Transport, WebRTC & SIP Trunking ($0.004 – $0.015 / min): Twilio, Telnyx, and Vonage (PSTN telephony connectivity) paired with LiveKit, Daily, or Agora (ultra-low-latency WebRTC edge routing, jitter buffering, VAD, and echo cancellation).
• Layer 2 — Real-Time Streaming STT ($0.004 – $0.012 / min): Deepgram Nova-3, AssemblyAI Universal-2, Gladia, Groq Whisper, Return Zero VITO, and CLOVA Speech. Even a 100ms delay in endpointing (detecting when the user stops speaking) degrades conversational naturalness.
• Layer 3 — Conversational Reasoning LLM ($0.005 – $0.040 / min): Low-TTFT (<250ms) models such as GPT-4o-mini, Claude 3.5 Haiku, and Gemini 2.0 Flash dominate chained pipelines. Alternatively, native Speech-to-Speech models (OpenAI Realtime API, Gemini Live, Kyutai Moshi) collapse Layers 2–4 into a single audio-token stream at roughly $0.06–$0.24 / min.
• Layer 4 — Ultra-Low-Latency Streaming TTS ($0.015 – $0.060 / min): Cartesia Sonic, ElevenLabs Flash v2.5, PlayHT, Rime AI, and OpenAI TTS.
• Layer 5 — Voice Agent Orchestration Platform ($0.05 – $0.09 / min base + provider pass-through): Vapi, Retell AI, Bland AI, and Synthflow manage barge-in handling, call recording, function calling (CRM/booking APIs), and silence detection—bringing the true all-in production cost to $0.09 – $0.22 / min.

---

4. Commercial APIs vs. Open-Weight Local Hosting: Breakeven & License Traps

Open-source audio models from China, the US, and Europe have closed the quality gap at breakneck speed, but code licenses (MIT/Apache) frequently differ from model weight licenses:

1. Commercially Permissive Open-Source Models (Apache 2.0 / MIT):

• Qwen3-TTS / CosyVoice 2 (Alibaba FunAudioLLM): Supports multilingual zero-shot cloning and emotional streaming synthesis across Chinese, English, Japanese, and Korean under Apache 2.0.
• GPT-SoVITS (RVC-Boss): Delivers industry-leading timbre cloning from 5–60 seconds of reference audio under the MIT License.
• Chatterbox (Resemble AI), F5-TTS & Kokoro-82M: MIT / Apache 2.0 weights capable of serving dozens of concurrent streams on a single RTX 4090 or L40S GPU.
• VOICEVOX (Hiho, Japan): While the engine is LGPL, each character voice bank carries its own commercial and attribution terms, requiring character-by-character verification.

2. Non-Commercial Weights (CC-BY-NC) Requiring Commercial Licensing:

• Fish Speech (Fish Audio): While the codebase is open-source, recent model weights use CC-BY-NC-SA 4.0 (Non-Commercial), requiring Fish Audio's commercial API or enterprise license for paid products.
• Meta AudioCraft (MusicGen / AudioGen) & Stability Stable Audio Open: Model weights are released under CC-BY-NC 4.0, meaning self-hosting these weights to sell generated music commercially violates the license (commercial users must use Stability Audio's paid API or enterprise licensing).

---

5. Deepfake Regulations, Voice Publicity Rights & Platform Guardrails

• Mandatory Voice Captcha & Publicity Right Laws: Under US FTC enforcement, Tennessee's ELVIS Act (effective 2024 — explicitly extending publicity rights to protect artists' voices from unauthorized AI cloning), and EU AI Act transparency rules, major platforms (ElevenLabs, OpenAI Audio, Supertone, Typecast) enforce live Voice Captcha verification for custom voice cloning alongside imperceptible audio watermarking (AudioSeal / C2PA).
• YouTube, TikTok & Spotify Labeling and Monetization Rules: YouTube requires creators to disclose "Altered or synthetic content" when realistic voices or faces are synthetically generated. Meanwhile, Spotify and Apple Music actively purge bot-streamed AI tracks and unauthorized artist vocal clones.
• No Commercial Rights on Free Tiers: Suno, Udio, ElevenLabs, and Typecast do not grant commercial monetization rights (YouTube monetization, paid ads, client delivery) to audio generated on Free plans. Always retain subscription receipts and generation timestamps when publishing commercial assets.

---

6. How to Use the Voice & Audio AI Market Map

In the interactive Voice & Audio AI Market Map, you can filter and sort all 99 players in real time:

• Category Filter (cat): ① TTS & Voice Cloning · ② Real-Time Voice Agents & STT · ③ AI Music & Sound Effects · ④ Video Dubbing & Localization (with secondary category toggle)
• Region Filter (region): US/Global · KR · JP · CN · EU
• Tier Filter (tier): Tier 1 · Tier 2 · Rising · Open Source
• Verification Badges (v) & Row Expansion: Click any row to inspect founding year, ownership, funding/valuation, latency, per-minute/monthly pricing, open-source license terms, and primary source links (src).

👉 Launch the Full Voice & Audio AI Market Map (88 Players)